Forest biomass nondestructive testing device and method in forestry carbon sink measurement

Abstract

The invention discloses a forest biomass nondestructive testing device and method in forestry carbon sink measurement, and relates to the technical field of forestry resource monitoring and carbon sink evaluation, the device comprises a multi-mode sensing unit, an underground biomass monitoring module and a data processing unit, the multi-mode sensing unit collects forest stand data through cooperation of a microwave radar, a laser radar and a hyperspectral imaging assembly. The underground biomass monitoring module collects acoustic impedance data of a root system; the data processing unit comprises an environment feature recognition module, a dynamic calibration engine, an organ recognition unit, a dynamic carbon distribution unit and a carbon sink metering module, the method comprises the following steps: cooperatively acquiring data through an air base and a foundation, analyzing environment features, dynamically calibrating, segmenting organs and generating a carbon distribution coefficient, and finally calculating the carbon sink amount of a single tree according to a formula. According to the invention, non-contact multi-parameter synchronous acquisition is realized, environmental changes are adapted through dynamic calibration, accurate metering is realized by means of organ-level carbon distribution, and the precision and reliability of forestry carbon sink metering are improved.

Description

Technical Field

[0001] This invention relates to the field of forestry resource monitoring and carbon sequestration assessment technology, specifically to a non-destructive testing device and method for forest biomass in forestry carbon sequestration measurement. Background Technology

[0002] Forestry carbon sequestration is a core technological step in accurately quantifying forest carbon storage, and its core requirement lies in efficiently acquiring forest biomass and carbon sequestration data. Traditional forest biomass measurement methods mainly rely on weighing felled trees or manually measuring diameter at breast height (DBH) and tree height, then converting the carbon storage using empirical biomass equations. These methods have significant limitations: firstly, weighing felled trees is destructive and can damage the forest ecosystem; secondly, manual measurement requires a large workforce, is time-consuming and inefficient, and is difficult to adapt to the needs of high-frequency, large-scale dynamic monitoring of forestry carbon sequestration. With the increasing awareness of ecological protection and the growing demand for more refined carbon sequestration monitoring, traditional methods can no longer meet the needs of practical applications.

[0003] In existing technologies, research on non-destructive testing of forest biomass has made some progress, mainly falling into two categories. One category is single-parameter dynamic monitoring devices, such as the tree growth monitoring device disclosed in the patent (CN222837497U). This device achieves continuous measurement of diameter at breast height (DBH) through a converging component and a damping wheel. However, this type of device can only acquire DBH data for a single tree and cannot simultaneously collect key biomass measurement parameters such as tree height and crown width. Furthermore, its mechanical structure is easily affected by complex weather conditions such as rain and fog, resulting in insufficient environmental adaptability. The other category is multi-source data acquisition devices, such as the device and method for assessing carbon storage in poplar plantations based on biomass conversion proposed in the patent (CN118470523A). This device estimates carbon storage by collecting leaf and branch samples and combining them with image modeling. While avoiding the destructive operation of felling trees, it still requires destructive sampling of leaves and branches. Moreover, it lacks a dynamic calibration mechanism for multi-sensor data, leading to significantly increased model errors (typically >15%) in complex forest stand environments such as foliage shading or changes in light intensity.

[0004] In current technological applications, two main problems urgently need to be addressed. Firstly, there is the issue of dynamic distortion in multimodal sensing data: in high-density forest stands, sensors such as optical, LiDAR, and microwave sensors are prone to systematic measurement biases due to differences in physical properties, such as missing point cloud data caused by foliage obstruction and spectral drift caused by rain and fog scattering. Existing calibration techniques often employ static calibration or single-sensor compensation, failing to dynamically adapt to real-time changes in environmental factors such as forest stand structure and meteorological conditions. Secondly, there is the problem of inefficient carbon allocation: traditional methods use a fixed carbon allocation coefficient (typically 0.45-0.55) for carbon storage estimation, neglecting the differences in carbon content among different tree species, ages, and forest organs (such as trunks, branches, and leaves). In reality, the carbon content of the xylem in tree trunks can reach 48-52%, while the carbon content of leaves is only 45-48%. This inefficient allocation method leads to an estimation error of 10-20% in the carbon sink of a single tree.

[0005] In summary, while existing technologies have made improvements in single-point parameter measurement or local data acquisition, they have not achieved an organic combination of non-contact, multi-parameter synchronous acquisition and dynamic environmental adaptive carbon allocation optimization. Especially in complex scenarios such as mixed forests and steep slopes, their measurement accuracy and carbon sink assessment reliability fail to meet the carbon trading market's certification requirement of <5% measurement error, thus failing to provide effective technical support for accurate measurement of forestry carbon sinks. Therefore, developing a non-destructive testing scheme that integrates multimodal sensing dynamic fusion and organ-level carbon allocation mechanisms has become a key requirement for improving the accuracy of forestry carbon sink measurement. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a non-destructive testing device and method for forest biomass in forestry carbon sequestration. By setting up a multimodal sensing unit, it utilizes microwave radar, lidar, and hyperspectral imaging components to collaboratively acquire three-dimensional point cloud and spectral data of the forest stand, combined with a groundwater biomass monitoring module to obtain root acoustic impedance data. A data processing unit is configured to analyze forest stand parameters through environmental feature identification; a dynamic calibration engine predicts deviations and allocates fusion weights based on a long short-term memory neural network; an organ identification unit segments individual tree organ regions; and a dynamic carbon allocation unit generates organ-level carbon allocation coefficients. Finally, the carbon sequestration of an individual tree is calculated according to a formula. This invention enables non-contact, accurate measurement of forest biomass, dynamically adapts to environmental changes, improves carbon sequestration accuracy, and overcomes the shortcomings of existing technologies, such as strong destructiveness, large errors, and coarse carbon allocation.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In one aspect, a non-destructive testing device for forest biomass in forestry carbon sequestration measurement, comprising:

[0008] The multimodal sensing unit includes a microwave radar component, a lidar component, and a hyperspectral imaging component. The microwave radar component emits multi-band penetrating waves and receives reflected signals. The lidar and hyperspectral imaging component synchronously acquire three-dimensional point cloud and spectral reflectance data through a common optical path lens.

[0009] The underground biomass monitoring module includes an array of acoustic wave transmitters and receiving sensors buried in the soil, which collect acoustic impedance data of the root region through directional acoustic waves.

[0010] A data processing unit, communicatively connected to the above-mentioned units, includes:

[0011] The environmental feature recognition module outputs parameters such as canopy closure, weather conditions, and tree species type based on laser point cloud porosity, microwave scattering intensity, and hyperspectral feature spectrum.

[0012] The dynamic calibration engine receives environmental parameters, predicts sensor deviations through a long short-term memory neural network, allocates fusion weights according to real-time standard errors, and generates a set of calibrated forest structure parameters including diameter at breast height, tree height, crown volume, and timber density.

[0013] The organ recognition unit, including an edge computing chip and a pre-trained U-Net3+ segmentation model, segments the trunk, branches, and leaf regions of a single tree based on calibration parameters.

[0014] The dynamic carbon allocation unit includes a near-infrared spectroscopy analysis module and a carbon allocation coefficient mapping table. The former collects the lignin content of the leaf spectrum, and the latter stores the correspondence between the lignin content and the carbon allocation coefficient.

[0015] The carbon sequestration metering module, with communication connection to the dynamic carbon allocation unit, is configured to operate by C. 单木 =∑ k (V k ×ρ k ×CF k The formula for calculating carbon sequestration per tree, C 单木 This represents the carbon sequestration of a single tree, where k represents the carbon sequestration of the trunk, branches, and leaves, and V represents the carbon sequestration of the tree. k Generated by the 3D point cloud segmentation mask of the organ recognition unit, ρ k Based on dynamic calibration engine data, CF k It is generated by querying the mapping table from the dynamic carbon allocation unit.

[0016] Furthermore, the microwave radar component includes:

[0017] An airborne scanning module, mounted on a fixed-wing UAV, is configured to transmit microwaves in the 0.5GHz, 1.2GHz, and 3.0GHz bands at a flight altitude of 150 meters to scan the upper structure of the forest canopy;

[0018] The ground-based scanning module, mounted on an unmanned vehicle platform, is configured to work in conjunction with lidar and hyperspectral imaging components to scan mid-level forest stands.

[0019] The data transmission interface between the airborne scanning module and the ground-based scanning module is connected to the data processing unit via a time-division multiplexing protocol.

[0020] Furthermore, the common-path lens for the lidar component and the hyperspectral imaging component includes:

[0021] The beam splitter is configured to divide the incident light into a visible-near-infrared channel and a laser channel according to wavelength.

[0022] The optical path end of the visible-near-infrared channel is connected to the photosensitive surface of the hyperspectral imaging sensor;

[0023] The optical path end of the laser channel is connected to the photosensitive surface of the lidar receiver;

[0024] The optical axis of the beam splitter coincides with the emission axis of the lidar.

[0025] Furthermore, the U-Net3+ segmentation model in the organ recognition unit further includes:

[0026] The attention mechanism branch is configured to enhance the weight ratio of trunk edge features in the feature map;

[0027] The multi-temporal learning module is configured to integrate historical time-series point cloud data with real-time acquired data and output organ segmentation masks.

[0028] Furthermore, the carbon allocation coefficient mapping table in the dynamic carbon allocation unit is constructed according to the following rules:

[0029] When the organ type is a tree trunk and the lignin content is between 28% and 32%, the carbon partition coefficient is between 0.50 and 0.52.

[0030] When the organ type is a leaf and the lignin content is in the range of 18% to 22%, the carbon partition coefficient is 0.46 to 0.48.

[0031] Furthermore, the underground biomass monitoring module includes:

[0032] A biomimetic root-shaped sensor housing encapsulates an acoustic wave transmitter and a receiving sensor inside.

[0033] The acoustic frequency modulation module is configured to switch the transmission frequency from 20kHz to 50kHz according to the soil type.

[0034] A root system architecture database stores spatial distribution models of root systems for typical tree species. It is configured to match acoustic impedance data with the distribution models and output root biomass.

[0035] Furthermore, the fusion weight allocation of the dynamic calibration engine is performed according to the following formula:

[0036]

[0037] W i This represents the fusion weight coefficient of the i-th sensor, used for weighted fusion of multimodal sensing data, dimensionless, σ i The standard deviation of the real-time error of the i-th sensor is represented by σ, which is generated by statistically analyzing historical deviations under the current environmental characteristics using a long short-term memory neural network. The unit is a physical quantity consistent with the original sensor data. n represents the total number of sensors participating in the fusion, including microwave radar components, lidar components, and hyperspectral imaging components. j Let j represent the standard deviation of the real-time error of the j-th sensor, where j = 1, 2, ..., n. This represents the sum of the reciprocals of the squares of the standard deviations of all sensor errors, used to normalize the weighting coefficients.

[0038] On the other hand, a non-destructive testing method for forest biomass in forestry carbon sequestration includes the following specific steps:

[0039] S100: Multi-band microwave scanning of the upper canopy is conducted via an airborne scanning module to generate canopy point cloud data;

[0040] S200: Simultaneously collects laser point cloud and hyperspectral data of mid-level forest stands through the ground-based scanning module;

[0041] S300: The environmental feature recognition module analyzes parameters such as canopy closure, weather conditions, and tree species type.

[0042] S400 predicts data deviation based on long short-term memory neural network through dynamic calibration engine, allocates fusion weight according to sensor error standard deviation, and outputs calibrated diameter at breast height, tree height, crown volume and wood density.

[0043] S500: The organ regions of a single tree, including its trunk, branches, and leaves, are segmented using an organ recognition unit.

[0044] S600: Obtain leaf lignin content through near-infrared spectroscopy analysis module, and generate organ-level carbon allocation coefficient by querying carbon allocation coefficient mapping table;

[0045] S700, press C 单木 =∑ k (V k ×ρ k ×CF k The formula is used to calculate the carbon sink of a single tree.

[0046] Compared with existing technologies, this non-destructive testing device and method for forest biomass in forestry carbon sequestration has the following advantages:

[0047] I. This invention addresses the destructive nature of traditional forest biomass measurement methods that rely on weighing felled trees. Through the collaborative design of a multimodal sensing unit and a dynamic calibration engine, it achieves non-destructive, high-precision biomass detection. Specifically, the multimodal sensing unit integrates microwave radar, lidar, and hyperspectral imaging components to simultaneously acquire three-dimensional point cloud and spectral reflectance data of trees, covering structural information at different levels of the aboveground parts. The underground biomass monitoring module acquires root acoustic impedance data using directional acoustic wave technology, overcoming the gap in traditional methods for monitoring underground parts. The dynamic calibration engine constructs an environment-sensor bias prediction model based on a long short-term memory neural network, and dynamically allocates data fusion weights using real-time standard error, effectively correcting systematic errors caused by environmental interference from multi-source sensors. This avoids the disturbance to the forest ecosystem caused by traditional destructive sampling. Furthermore, through multi-dimensional data fusion and intelligent calibration, the measurement error of single-tree biomass parameters is reduced to within 5%, significantly improving the continuity and reliability of carbon sequestration measurement.

[0048] II. This invention addresses the problem of coarse-grained bias caused by the use of fixed carbon allocation coefficients in traditional carbon sink estimation. By integrating an organ identification unit with a near-infrared spectroscopy analysis module, it achieves dynamic generation of organ-level carbon allocation coefficients. The organ identification unit adopts an improved U-Net3+ segmentation model, introduces an attention mechanism branch to strengthen the weight of trunk edge features, and integrates multi-temporal point cloud data to optimize the segmentation results, accurately distinguishing organ regions such as trunks, branches, and leaves. The near-infrared spectroscopy analysis module extracts the spectral characteristics of leaf lignin content and, combined with a pre-stored carbon allocation coefficient mapping table, establishes a quantitative correlation between organ lignin content and carbon allocation coefficient. This overcomes the limitations of traditional methods that ignore differences in tree species, age, and organs. Through precise calibration of organ-level carbon allocation coefficients, the estimation deviation of single-tree carbon sink is reduced from 10-20% to less than 3%, significantly improving the precision of carbon sink measurement and providing more scientific data support for forestry carbon sink trading and ecological compensation.

[0049] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0051] Figure 1 This is a flowchart illustrating the operation of the present invention;

[0052] Figure 2 This is a structural diagram of the modal sensing and data acquisition of the present invention;

[0053] Figure 3 This is a block diagram of the device structure of the present invention. Detailed Implementation

[0054] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0055] Example 1

[0056] like Figures 1 to 3 As shown in this embodiment, a non-destructive testing device and method for forest biomass in forestry carbon sequestration is disclosed. The device acquires three-dimensional structural and spectral information of different forest stand levels through a multimodal sensing unit, obtains root biomass data by combining it with a subsurface biomass monitoring module, and completes environmental parameter analysis, dynamic data calibration, organ region segmentation, carbon allocation coefficient generation, and carbon sequestration calculation through a data processing unit. This method, through steps such as airborne and ground-based collaborative acquisition, dynamic weighted fusion calibration, organ-level parameter segmentation, and precise carbon allocation, achieves non-contact, high-precision measurement of forest biomass. It effectively solves the problems of strong destructiveness, low efficiency, and large errors in traditional methods, providing reliable technical support for accurate measurement of forestry carbon sequestration.

[0057] The overall architecture and collaborative working mechanism of the device are implemented as follows;

[0058] The non-destructive testing device for forest biomass in forestry carbon sequestration in this embodiment mainly consists of three parts: a multimodal sensing unit, an underground biomass monitoring module, and a data processing unit. Each part achieves data interaction and collaborative work through wired or wireless communication to form a complete non-destructive testing system.

[0059] The multimodal sensing unit, as the core component of data acquisition, is responsible for acquiring information on the aboveground structure and spectral information of the forest stand. Through the coordinated operation of microwave radar, lidar, and hyperspectral imaging components, it achieves simultaneous acquisition of data across different frequency bands and dimensions. The underground biomass monitoring module focuses on monitoring the root biomass of trees. It collects physical characteristic data of the root region through sensors buried in the soil, compensating for the lack of measurement of underground components in traditional methods. The data processing unit receives the raw data from the two modules and, after a series of analysis, calibration, segmentation, and calculation processes, finally outputs the carbon sequestration results for individual trees.

[0060] In practical applications, the multimodal sensing unit and the underground biomass monitoring module start working synchronously according to a preset collection cycle, and the collected raw data is transmitted to the data processing unit in real time. The data processing unit automatically assigns the data to the corresponding processing module according to the data type, and after hierarchical processing, it forms a structured measurement result. The entire process does not require felling trees or destructive sampling, achieving truly non-destructive testing.

[0061] The structure of the multimodal sensing unit and the implementation of the data acquisition process are as follows;

[0062] The multimodal sensing unit includes a microwave radar component, a lidar component, and a hyperspectral imaging component. The three components achieve data acquisition synchronization through unified timing control, ensuring the accuracy of subsequent data fusion.

[0063] The microwave radar assembly consists of an airborne scanning module and a ground-based scanning module, which collect data on the upper and middle canopy stands, respectively. The airborne scanning module is mounted on a fixed-wing UAV and cruises along a planned route at a preset flight altitude. Its emitted multi-band microwaves can penetrate the canopy surface to acquire three-dimensional structural information of the upper canopy. The different frequency bands of microwaves complement each other in terms of penetration capability and resolution, adapting to stand environments with varying canopy closures. The ground-based scanning module is mounted on an unmanned vehicle platform and moves along the stand transect on the ground, maintaining spatial coordination with the lidar and hyperspectral imaging components to jointly collect structural and spectral data of the middle canopy stands. The airborne and ground-based scanning modules establish a communication connection with the data processing unit via a time-division multiplexing protocol, avoiding data transmission conflicts and ensuring the complete upload of raw data.

[0064] The lidar and hyperspectral imaging components achieve synchronous data acquisition through a common-path lens, which includes a beam-splitting prism, a visible-near-infrared channel, and a laser channel. The beam-splitting prism splits the incident light according to its wavelength characteristics. The visible-near-infrared light is guided into the visible-near-infrared channel and ultimately projected onto the photosensitive surface of the hyperspectral imaging sensor, forming hyperspectral reflectance data. The laser light is guided into the laser channel and projected onto the photosensitive surface of the lidar receiver, forming 3D point cloud data. The optical axis of the beam-splitting prism is aligned with the lidar's emission axis, ensuring consistency between the lidar's emission and reception directions and reducing measurement errors caused by optical path deviations. During acquisition, the lidar component emits laser pulses and receives reflected signals to calculate the target's distance information and generate a high-precision 3D point cloud. The hyperspectral imaging component simultaneously records the target's reflectivity at different wavelengths, forming spectral characteristic data. The data acquisition frequencies of both components are kept consistent, achieving precise spatial matching.

[0065] The structure of the underground biomass monitoring module and the implementation of the root data acquisition process are as follows;

[0066] The underground biomass monitoring module is used to obtain biomass information of forest tree roots. Its structure includes a biomimetic root-shaped sensor housing, an acoustic frequency modulation module, and a root configuration database.

[0067] The biomimetic root-shaped sensor housing adopts a shape similar to natural roots, allowing it to be buried in root distribution areas without damaging the soil structure. The internally encapsulated sound wave transmitter and receiver maintain a fixed spatial position, ensuring the directionality of sound wave transmission and reception. Under the control of the sound wave frequency modulation module, the sound wave transmitter emits directional sound waves. As the sound waves propagate through the soil, they are reflected and refracted upon encountering the roots. The receiver captures the reflected sound signals and converts them into electrical signals, forming acoustic impedance data.

[0068] The acoustic frequency modulation module automatically adjusts the transmission frequency according to different soil types. Different types of soil (such as clay, loam, and sand) exhibit different propagation attenuation characteristics for sound waves. By switching the transmission frequency, the sound waves can maintain optimal propagation in the target soil environment, improving the signal-to-noise ratio of the acoustic impedance data. The collected acoustic impedance data is transmitted to a root system configuration database for matching analysis. This database stores spatial distribution models of root systems for various typical tree species, including characteristic parameters such as root diameter, depth, and density distribution. By comparing the actual collected acoustic impedance data with the models in the database, root biomass information can be retrieved, enabling the quantitative measurement of underground biomass.

[0069] The module functions of the data processing unit and the implementation of the data processing process are as follows;

[0070] The data processing unit is the core component for data parsing, calibration, and carbon sink calculation. It includes an environmental feature identification module, a dynamic calibration engine, an organ identification unit, a dynamic carbon allocation unit, and a carbon sink measurement module. Each module works together sequentially according to the data processing flow.

[0071] The environmental feature recognition module receives laser point cloud data, microwave scattering data, and hyperspectral feature spectrum data from the multimodal sensing unit. It analyzes the canopy closure by calculating the porosity of the laser point cloud; a higher porosity indicates lower canopy closure. Based on the changing trends of microwave scattering intensity and the changes in reflectance in specific bands of the hyperspectral feature spectrum, it determines the current weather conditions. By extracting tree-specific spectral fingerprints from the hyperspectral feature spectrum and combining them with a known tree species spectral database, it identifies the tree species types within the forest stand. The analyzed canopy closure, weather conditions, and tree species type parameters serve as environmental feature parameters, providing a basis for subsequent data calibration.

[0072] The dynamic calibration engine calibrates the raw sensor data based on environmental characteristic parameters, where the fusion weight allocation is performed according to the following formula:

[0073]

[0074] W i This represents the fusion weight coefficient of the i-th sensor, used for weighted fusion of multimodal sensing data, dimensionless, σ i The standard deviation of the real-time error of the i-th sensor is represented by σ, which is generated by statistically analyzing historical deviations under the current environmental characteristics using a long short-term memory neural network. The unit is a physical quantity consistent with the original sensor data. n represents the total number of sensors participating in the fusion, including microwave radar components, lidar components, and hyperspectral imaging components. j Let j represent the standard deviation of the real-time error of the j-th sensor, where j = 1, 2, ..., n. This represents the sum of the reciprocals of the squares of the standard deviations of all sensor errors, used to normalize the weighting coefficients.

[0075] By learning from historical data using a Long Short-Term Memory (LSTM) neural network, this network can capture the temporal correlation between environmental factors and sensor biases, thereby predicting the bias of each sensor under current environmental characteristics. Subsequently, fusion weights are calculated based on the real-time standard errors of each sensor. The formula for calculating the fusion weight is the ratio of the reciprocal of the square of the standard deviation of each sensor's error to the sum of the reciprocals of the squares of the standard deviations of all sensor errors. Sensors with smaller standard deviations have larger fusion weights and occupy a higher proportion of weight in data fusion. Through this dynamic weight allocation method, the raw data from microwave radar, lidar, and hyperspectral imaging are weighted and fused to ultimately generate calibrated forest structure parameters such as diameter at breast height (DBH), tree height, crown volume, and timber density, effectively reducing systematic errors caused by environmental factors.

[0076] The organ recognition unit uses a pre-trained U-Net3+ segmentation model to process the calibrated 3D point cloud data. This model includes an attention mechanism branch and a multi-temporal learning module. The attention mechanism branch improves the accuracy of trunk region segmentation by strengthening the weight of trunk edge features in the feature map. The multi-temporal learning module fuses historical time-series point cloud data with real-time acquired data, and optimizes the segmentation results by utilizing the continuous characteristics of tree growth. Finally, it outputs the segmentation masks of the trunk, branches, and leaf regions of individual trees, laying the foundation for subsequent organ-level parameter calculations.

[0077] The dynamic carbon allocation unit consists of a near-infrared spectroscopy analysis module and a carbon allocation coefficient mapping table. The near-infrared spectroscopy analysis module analyzes leaf spectral data acquired by the hyperspectral imaging component, extracts spectral features related to lignin content, and calculates the lignin content of the leaves. The carbon allocation coefficient mapping table stores the relationship between carbon allocation coefficients and lignin content for different organ types. Based on the organ type determined by the organ identification unit and the lignin content obtained by the near-infrared spectroscopy analysis module, the corresponding carbon allocation coefficient can be retrieved from the mapping table, achieving precise carbon allocation at the organ level.

[0078] The specific implementation steps of the non-destructive testing method for forest biomass are as follows;

[0079] The non-destructive testing method for forest biomass based on the above-mentioned device shall be performed according to the following steps:

[0080] First, the airborne scanning module is activated. A fixed-wing UAV equipped with this module flies above the forest stand along a preset route, emitting multi-band microwaves to scan the upper canopy. After penetrating the canopy layer, the microwaves are reflected by different layers of tree structure. The airborne scanning module receives the reflected signals and converts them into electrical signals. After preprocessing, canopy point cloud data is generated, completing the acquisition of information on the upper canopy structure.

[0081] Secondly, while performing airborne scanning, the ground-based scanning module, lidar component, and hyperspectral imaging component are activated. The unmanned vehicle equipped with these components moves along a pre-set transect on the forest stand ground. The ground-based scanning module emits microwaves to acquire structural data of the mid-level forest stand; the lidar component emits laser pulses and receives reflected signals through the laser channel of the common-path lens to generate three-dimensional point cloud data of the mid-level forest stand; the hyperspectral imaging component receives reflected light through the visible-near-infrared channel of the common-path lens to generate hyperspectral data, achieving simultaneous acquisition of mid-level forest stand structure and spectral data.

[0082] Next, the environmental feature recognition module analyzes the collected laser point cloud data, microwave scattering data, and hyperspectral data. It calculates the canopy closure by the porosity of the laser point cloud, determines the weather condition based on changes in microwave scattering intensity and hyperspectral characteristic spectra, identifies tree species by matching the hyperspectral characteristic spectra with a tree species spectral database, and outputs environmental feature parameters.

[0083] Then, the dynamic calibration engine receives environmental feature parameters, calls the long short-term memory neural network to predict the deviation of each sensor in the current environment, calculates the real-time standard error of each sensor, assigns weights according to the fusion weight formula, performs weighted fusion on the original sensor data, and generates calibrated forest structure parameters such as diameter at breast height, tree height, crown volume and wood density.

[0084] Subsequently, the organ recognition unit loads the calibrated 3D point cloud data, strengthens the trunk edge features through the attention mechanism branch of the U-Net3+ segmentation model, and combines historical data with the multi-temporal learning module to segment the trunk, branches, and leaf regions of a single tree, outputting the segmentation mask of each organ.

[0085] Next, the near-infrared spectroscopy analysis module analyzes the hyperspectral data of the leaves and calculates the lignin content of the leaves; the dynamic carbon allocation unit determines the organ type based on the organ identification results, and generates the carbon allocation coefficient corresponding to each organ by querying the carbon allocation coefficient mapping table in combination with the lignin content.

[0086] Finally, the carbon sequestration metering module calls the organ segmentation masks output by the organ recognition unit to generate volume parameters, combines them with the density parameters obtained by the dynamic calibration engine and the carbon allocation coefficient obtained by the dynamic carbon allocation unit, and calculates the volume parameters according to C. 单木 =∑ k (V k ×ρ k ×CF k The formula for calculating carbon sequestration per tree, C 单木 This represents the carbon sequestration of a single tree, where k represents the carbon sequestration of the trunk, branches, and leaves, and V represents the carbon sequestration of the tree. k Generated by the 3D point cloud segmentation mask of the organ recognition unit, ρ k Based on dynamic calibration engine data, CF k The carbon sink of a single tree is obtained by querying the mapping table through the dynamic carbon allocation unit.

[0087] In summary, this embodiment achieves synchronous acquisition of data from different levels and dimensions of the aboveground parts of the forest stand through a multimodal sensing unit. Combined with the underground biomass monitoring module, it fills the gap in underground root biomass measurement. Through the collaborative work of multiple modules in the data processing unit, it realizes end-to-end processing from raw data to carbon sequestration results. Specifically, the dynamic calibration engine, through a long short-term memory neural network and dynamic weight allocation mechanism, effectively reduces measurement errors caused by environmental factors; the U-Net3+ model of the organ identification unit achieves accurate segmentation of organ regions; and the dynamic carbon allocation unit, based on a carbon allocation coefficient mapping table of lignin content, improves the precision of carbon sequestration calculation.

[0088] This device and method completely eliminate the destructive operation of traditional methods, significantly improve data acquisition efficiency and measurement accuracy, can adapt to different forest stand environments and weather conditions, and meet the high-frequency, large-scale, and high-precision monitoring needs of forestry carbon sequestration. It has important practical value and significance for promotion.

[0089] Example 2

[0090] like Figure 1 As shown in Example 1, this example provides a detailed description of the workflow of the non-destructive testing device for forest biomass in forestry carbon sequestration. This workflow achieves high-precision non-destructive measurement of forest biomass through multi-platform collaborative data acquisition, multi-modal data fusion, and organ-level carbon allocation. Specifically, it includes the following steps:

[0091] 1. Aerial Data Acquisition: A fixed-wing UAV equipped with an aerial scanning module for microwave radar takes off to an altitude of 150 meters and cruises along a preset route. The aerial scanning module sequentially transmits microwave signals in the 0.5GHz, 1.2GHz, and 3.0GHz frequency bands, penetrating the upper canopy structure and receiving reflected echoes. After signal analysis, it generates canopy point cloud data and transmits it back to the data processing unit in real time via a wireless transmission link.

[0092] 2. Ground-based Coordinated Data Acquisition: An unmanned vehicle platform equipped with a microwave radar ground-based scanning module, a lidar component, and a hyperspectral imaging component is simultaneously activated and travels at a constant speed along the forest transect. The ground-based scanning module emits microwaves to scan the mid-level forest stand structure; the lidar component acquires three-dimensional point cloud data through the laser channel in the common-path lens; and the hyperspectral imaging component simultaneously acquires spectral reflectance data through the visible-near-infrared channels. The data from all three sources are consistent in time and space.

[0093] 3. Underground data acquisition: The biomimetic root-shaped acoustic wave sensor array buried in the root distribution area starts working. The acoustic wave frequency modulation module adaptively selects a transmission frequency of 20–50kHz according to the soil moisture and type, directionally transmits sound waves and receives reflected signals, obtains acoustic impedance data of the root area, and transmits it to the data processing unit.

[0094] 4. Environmental Feature Analysis: The environmental feature identification module calculates the canopy porosity and analyzes the stand canopy closure based on the received laser point cloud data; it analyzes the microwave scattering intensity fluctuations and the reflectance changes in water vapor absorption bands (such as 940nm and 1130nm) in the hyperspectral feature spectrum to determine the weather status (sunny, rainy, foggy); and it matches the hyperspectral features with the tree species spectral library to identify the main tree species types.

[0095] 5. Dynamic calibration of multi-source data: The dynamic calibration engine receives environmental parameters, calls a pre-trained long short-term memory neural network (LSTM), predicts the measurement deviation of each sensor (microwave radar, lidar, hyperspectral) in the current environment based on historical data, calculates the real-time error standard deviation of each sensor, performs weighted fusion according to the weight allocation formula, and outputs the calibrated data of tree diameter at breast height, tree height, crown volume and wood density.

[0096] 6. Organ point cloud segmentation: The organ recognition unit loads the calibrated 3D point cloud data, strengthens the extraction of tree trunk edge features by embedding the U-Net3+ model with an attention mechanism branch, and combines the historical time series point cloud with the multi-temporal learning module to achieve accurate segmentation of the trunk, branches and leaf regions of a single tree, and generate volume masks for each organ.

[0097] 7. Carbon allocation coefficient generation: The near-infrared spectroscopy analysis module extracts characteristic bands related to lignin (such as 1680nm and 1730nm) from the leaf spectral data and calculates the leaf lignin content; the dynamic carbon allocation unit queries the carbon allocation coefficient mapping table based on organ type (trunk, branch, leaf) and lignin content to generate organ-level carbon allocation coefficients.

[0098] 8. Carbon Sequestration Calculation and Output: The carbon sequestration metering module calculates the volume V of each organ based on the organ segmentation mask. k Combined with the calibrated wood density ρ k and carbon allocation coefficient CF k According to formula C 单木 =∑ k (V k ×ρ k ×CF k The carbon sequestration of a single tree is calculated, and the results are output to a display terminal or database, thus completing the non-destructive measurement of carbon sequestration of a single tree biomass.

[0099] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A non-destructive testing device for forest biomass in forestry carbon sequestration measurement, characterized in that, include: The multimodal sensing unit includes a microwave radar component, a lidar component, and a hyperspectral imaging component. The microwave radar component emits multi-band penetrating waves and receives reflected signals. The lidar and hyperspectral imaging component synchronously acquire three-dimensional point cloud and spectral reflectance data through a common optical path lens. The underground biomass monitoring module includes an array of acoustic wave transmitters and receiving sensors buried in the soil, which collect acoustic impedance data of the root region through directional acoustic waves. A data processing unit, communicatively connected to the above-mentioned units, includes: The environmental feature recognition module outputs parameters such as canopy closure, weather conditions, and tree species type based on laser point cloud porosity, microwave scattering intensity, and hyperspectral feature spectrum. The dynamic calibration engine receives environmental parameters, predicts sensor deviations through a long short-term memory neural network, allocates fusion weights according to real-time standard errors, and generates a set of calibrated forest structure parameters including diameter at breast height, tree height, crown volume, and timber density. The organ recognition unit, including an edge computing chip and a pre-trained U-Net3+ segmentation model, segments the trunk, branches, and leaf regions of a single tree based on calibration parameters. The dynamic carbon allocation unit includes a near-infrared spectroscopy analysis module and a carbon allocation coefficient mapping table. The former collects the lignin content of the leaf spectrum, and the latter stores the correspondence between the lignin content and the carbon allocation coefficient. The carbon sequestration metering module, with communication connection to the dynamic carbon allocation unit, is configured to operate by C. 单木 =∑ k (V k ×ρ k ×CF k The formula for calculating carbon sequestration per tree, C 单木 This represents the carbon sequestration of a single tree, where k represents the carbon sequestration of the trunk, branches, and leaves, and V represents the carbon sequestration of the tree. k Generated by the 3D point cloud segmentation mask of the organ recognition unit, ρ k Based on dynamic calibration engine data, CF k It is generated by querying the mapping table from the dynamic carbon allocation unit.

2. The non-destructive testing device for forest biomass in forestry carbon sequestration according to claim 1, characterized in that, The microwave radar component includes: An airborne scanning module, mounted on a fixed-wing UAV, is configured to transmit microwaves in the 0.5GHz, 1.2GHz, and 3.0GHz bands at a flight altitude of 150 meters to scan the upper structure of the forest canopy; The ground-based scanning module, mounted on an unmanned vehicle platform, is configured to work in conjunction with lidar and hyperspectral imaging components to scan mid-level forest stands. The data transmission interface between the airborne scanning module and the ground-based scanning module is connected to the data processing unit via a time-division multiplexing protocol.

3. The non-destructive testing device for forest biomass in forestry carbon sequestration according to claim 1, characterized in that, The common-path lens for the lidar component and the hyperspectral imaging component includes: The beam splitter is configured to divide the incident light into a visible-near-infrared channel and a laser channel according to wavelength. The optical path end of the visible-near-infrared channel is connected to the photosensitive surface of the hyperspectral imaging sensor; The optical path end of the laser channel is connected to the photosensitive surface of the lidar receiver; The optical axis of the beam splitter coincides with the emission axis of the lidar.

4. The non-destructive testing device for forest biomass in forestry carbon sequestration according to claim 1, characterized in that, The U-Net3+ segmentation model in the organ recognition unit further includes: The attention mechanism branch is configured to enhance the weight ratio of trunk edge features in the feature map; The multi-temporal learning module is configured to integrate historical time-series point cloud data with real-time acquired data and output organ segmentation masks.

5. The non-destructive testing device for forest biomass in forestry carbon sequestration according to claim 1, characterized in that, The carbon allocation coefficient mapping table in the dynamic carbon allocation unit is constructed according to the following rules: When the organ type is a tree trunk and the lignin content is between 28% and 32%, the carbon partition coefficient is between 0.50 and 0.

52. When the organ type is a leaf and the lignin content is in the range of 18% to 22%, the carbon partition coefficient is 0.46 to 0.

48.

6. The non-destructive testing device for forest biomass in forestry carbon sequestration according to claim 1, characterized in that, The underground biomass monitoring module includes: A biomimetic root-shaped sensor housing encapsulates an acoustic wave transmitter and a receiving sensor inside. The acoustic frequency modulation module is configured to switch the transmission frequency from 20kHz to 50kHz according to the soil type. A root system architecture database stores spatial distribution models of root systems for typical tree species. It is configured to match acoustic impedance data with the distribution models and output root biomass.

7. The non-destructive testing device for forest biomass in forestry carbon sequestration according to claim 1, characterized in that, The fusion weight allocation of the dynamic calibration engine is performed according to the following formula: W i This represents the fusion weight coefficient of the i-th sensor, used for weighted fusion of multimodal sensing data. It is dimensionless, and σ i The standard deviation of the real-time error of the i-th sensor is represented by σ, which is generated by statistically analyzing historical deviations under the current environmental characteristics using a long short-term memory neural network. The unit is a physical quantity consistent with the original sensor data. n represents the total number of sensors participating in the fusion, including microwave radar components, lidar components, and hyperspectral imaging components. j Let j represent the real-time error standard deviation of the j-th sensor, where j = 1, 2, ..., n. This represents the sum of the reciprocals of the squares of the standard deviations of all sensor errors, used to normalize the weighting coefficients.

8. A non-destructive testing method for forest biomass in forestry carbon sequestration, applicable to the non-destructive testing device for forest biomass in forestry carbon sequestration as described in any one of claims 1-7, characterized in that, The specific steps of this method are as follows: S100: Multi-band microwave scanning of the upper canopy is conducted via an airborne scanning module to generate canopy point cloud data; S200: Simultaneously collects laser point cloud and hyperspectral data of mid-level forest stands through the ground-based scanning module; S300: The environmental feature recognition module analyzes parameters such as canopy closure, weather conditions, and tree species type. S400 predicts data deviation based on long short-term memory neural network through dynamic calibration engine, allocates fusion weight according to sensor error standard deviation, and outputs calibrated diameter at breast height, tree height, crown volume and wood density. S500: The organ regions of a single tree, including its trunk, branches, and leaves, are segmented using an organ recognition unit. S600: Obtain leaf lignin content through near-infrared spectroscopy analysis module, and generate organ-level carbon allocation coefficient by querying carbon allocation coefficient mapping table; S700, press C 单木 =∑ k (V k ×ρ k ×CF k The formula is used to calculate the carbon sink of a single tree.

Images