Hand-held forest region sample plot calibration surveying and mapping device and method

Abstract

The invention discloses a handheld forest region sample plot calibration surveying and mapping device and method, and relates to the technical field of forest resource investigation, and the device comprises a laser radar, a visual camera and a global navigation satellite system / inertial orientation positioning navigation system integrated navigation unit; when a user carries the device to move, the processing unit constructs a three-dimensional point cloud map and calculates a six-degree-of-freedom pose; when the signal of the global navigation satellite system is unavailable, calculating a current position by combining the three-dimensional point cloud map, the six-degree-of-freedom pose and inertial data; and the processing unit calculates parameters of the stumpage and the candidate area and generates a sample plot calibration surveying and mapping report. According to the invention, the device achieves the synchronous capturing of the characteristics of the trunk base and the canopy at a handheld height, an operator only needs to click and select an initial position on a touch screen, and the system combines laser instant positioning and map construction with a global navigation satellite system / inertial orientation positioning navigation system for tight coupling positioning. And automatically generating sample plot boundaries and ecological parameters conforming to regulations.

Description

Technical Field

[0001] This application relates to the field of forest resource survey technology, and in particular to a handheld forest area transect location and mapping device and method. Background Technology

[0002] Currently, global forestry resource monitoring is facing an urgent need for digital transformation and upgrading. According to relevant reports, traditional manual survey methods can no longer meet the needs of high-frequency, high-precision forest carbon sink measurement and biodiversity conservation. Practical needs also require the establishment of an integrated "sky-ground" monitoring system for forest resources, but existing technologies still have significant shortcomings in grassroots forestry surveys: traditional compass and tape measure measurements rely on manual operation, requiring a team of 4 people (100m×100m) to complete basic data collection for a single standard plot (2 hours), with a measurement error rate as high as 12%, and GPS positioning loss in complex terrain causing plot boundary offsets of more than 15 meters; emerging UAV aerial surveys can improve efficiency, but the cost of a single operation exceeds 5,000 yuan, and they cannot penetrate the dense canopy to obtain understory vegetation data, posing flight safety hazards in mountainous areas with slopes greater than 35°; commercial backpack laser scanning equipment can obtain 3D point clouds, but it weighs more than 15 kg and has a flight time of less than 4 hours, which cannot meet the all-weather operation needs of remote forest areas.

[0003] Against this backdrop, breakthroughs in micro-sensor technology and edge computing chips have made technological innovation possible. Velodyne's Puck Lite lidar module weighs only 830 grams, has an angular resolution of 0.1°, and, when paired with the NVIDIA Jetson AGX Orin chip, can process 200,000 point cloud data points per second in real time. Advances in visual-inertial odometry have enabled positioning accuracy to reach 0.3% range error in environments without a global navigation satellite system. However, existing equipment integration solutions mostly employ fixed sensor layouts, leading to spatiotemporal mismatch of multi-source data due to human movement during handheld operations. Point cloud and image registration errors exceed 5cm, failing to meet the centimeter-level accuracy requirements of forestry surveys. Furthermore, traditional surveying equipment lacks human-computer interaction design, still requiring professionals to operate complex post-processing software, resulting in a severe disconnect between the technical capabilities of grassroots forestry workers.

[0004] Current forestry plot survey techniques suffer from three major shortcomings: First, the measuring tools are primitive. Traditional compasses are susceptible to geomagnetic anomalies, leading to orientation errors exceeding ±5°. Measuring tape measures in dense shrubland terrain suffers from systematic errors exceeding 10% due to the difficulty of manual straightening. Boundary calibration of individual standard plots requires repeated corrections, taking up to two hours. Second, electronic equipment has poor environmental adaptability. While UAV aerial surveys can quickly acquire canopy data, the image overlap rate in dense forest areas is less than 60%, and the understory vegetation information loss rate exceeds 80%, failing to meet the complete requirements of carbon sequestration for stand vertical structure parameters. Backpack-mounted laser scanning systems weigh over 15 kg, posing a risk of slippage when operating on slopes exceeding 35°. Third, data processing is lagging. Existing mobile surveying equipment mostly adopts a step-by-step "acquisition-export-post-processing" model. Grassroots technicians need to master specialized point cloud processing software, resulting in a cycle of several weeks from fieldwork to report generation, which is ill-suited to the timeliness requirements of dynamic forest resource monitoring. Summary of the Invention

[0005] This application provides a handheld forest area transect mapping device and method to solve the problems of primitive forestry resource survey techniques and methods, poor environmental adaptability of electronic equipment, and lagging data processing in the prior art.

[0006] On the one hand, embodiments of this application provide a handheld forest area sample point location and mapping device, including: shell; The lidar, visual camera, and global navigation satellite system / inertial orientation and positioning navigation system integrated navigation unit are all located on the top outer part of the housing; The processing unit is located inside the casing; When the user moves with the device, the lidar collects point cloud data, and the processing unit constructs a 3D point cloud map based on the point cloud data; the visual camera collects stereo images of the surrounding environment, and the processing unit extracts feature points from the stereo images, combining the feature points and the device's inertial data to calculate the six-degree-of-freedom pose; when the Global Navigation Satellite System (GNSS) signal is available, the GNSS / Inertial Orientation and Positioning Navigation System integrated navigation unit collects GNSS positioning signals, and the processing unit combines the GNSS positioning signals, the 3D point cloud map, and the six-degree-of-freedom pose to calculate the current position; when the GNSS signal is unavailable, the GNSS / Inertial Orientation and Positioning Navigation System integrated navigation unit collects inertial data, and the processing unit combines the 3D point cloud map, the six-degree-of-freedom pose, and the inertial data to calculate the current position. After the user selects the starting point and direction of the sample plot, the processing unit analyzes the point cloud density in the 3D point cloud map and generates candidate regions. The starting point of the sample plot is located in the candidate regions. The processing unit analyzes the 3D point cloud map in the candidate regions, identifies standing trees, and uses a genetic algorithm to optimize the boundaries of the candidate regions. The processing unit calculates the parameters of the standing trees and the candidate regions after boundary optimization, and generates a sample plot mapping report based on the calculated parameters.

[0007] In one possible implementation, when the processing unit calculates the current position by combining the Global Navigation Satellite System (GNSS) positioning signal, the 3D point cloud map, and the six-DOF pose, the weights of the GNSS positioning signal, the 3D point cloud map, and the six-DOF pose are 60%, 30%, and 10%, respectively. When the processing unit calculates the current position by combining the 3D point cloud map, the six-DOF pose, and the inertial data, the weights of the 3D point cloud map, the six-DOF pose, and the inertial data are 70%, 25%, and 5%, respectively.

[0008] In one possible implementation, a clamp is placed between the lidar and the vision camera, such that the angle between the optical axes of the lidar and the vision camera is 12°.

[0009] In one possible implementation, the housing contains a memory unit, a storage unit, and a multi-sensor synchronization controller, while the outer surface of the housing contains a touch screen, buttons, a battery compartment, a data interface, and a display interface.

[0010] In one possible implementation, the processing unit uses a density-based clustering algorithm to identify tree trunk point cloud clusters in a 3D point cloud map, removes shrub layers and fallen trees to determine the number of standing trees in the candidate region. If the number of standing trees is insufficient, the processing unit expands the boundary of the candidate region.

[0011] In one possible implementation, after the processing unit completes the boundary optimization of the candidate region, the user can fine-tune the candidate region by dragging the boundary points of the candidate region. The processing unit performs cylindrical fitting on the standing point cloud in the fine-tuned candidate region. If the result of the cylindrical fitting exceeds the set threshold, the processing unit will generate a recalibration prompt.

[0012] In one possible implementation, before calculating the parameters of the candidate region, the processing unit preprocesses the 3D point cloud map: statistically filters out outliers in the 3D point cloud map; uses a cloth-simulated filtering algorithm to separate the ground point cloud; and uses a region growing method to segment individual trees.

[0013] In one possible implementation, the parameters include the diameter at breast height (DBH) of the standing trees, tree height, crown width, and canopy closure of the candidate area.

[0014] In one possible implementation, the sample location mapping report includes boundary coordinates, a 3D point cloud bird's-eye view and side view, a parameter statistics table, and a data quality assessment matrix.

[0015] On the other hand, this application also provides a handheld forest area transect mapping method, including: When the device moves, it acquires point cloud data and constructs a 3D point cloud map based on the point cloud data; Acquire stereo images of the surrounding environment, extract feature points from the stereo images, and calculate the six-degree-of-freedom pose by combining the feature points and inertial data; When the Global Navigation Satellite System (GNSS) signal is available, the system collects the GNSS positioning signal and calculates the current position by combining the GNSS positioning signal, a 3D point cloud map, and a six-degree-of-freedom pose. When the GNSS signal is unavailable, the system collects inertial data and calculates the current position by combining the 3D point cloud map, a six-degree-of-freedom pose, and inertial data. Obtain the starting point and direction of the sample plot selected by the user; Analyze the point cloud density in the 3D point cloud map to generate candidate regions, with the starting point of the sample plot located within the candidate regions; Analyze the 3D point cloud map in the candidate region, identify standing trees, and use a genetic algorithm to optimize the boundary of the candidate region; The parameters of the standing trees and the candidate areas after boundary optimization are calculated, and a sample location mapping report is generated based on the calculated parameters.

[0016] The handheld forest area transect surveying and mapping device and method described in this application have the following advantages: To address the issue of low accuracy in traditional measuring tools, this application utilizes multimodal sensor spatiotemporal synchronization technology to achieve deep fusion of a LiDAR (range accuracy ±2cm), a visual camera (angular resolution 0.1°), and a combined navigation unit of a Global Navigation Satellite System / Inertial Orientation and Positioning Navigation System (tightly coupled positioning accuracy ±0.3m). After the user interactively selects the starting point of the sample plot, the system automatically completes the sample plot boundary calibration and 3D point cloud modeling with centimeter-level accuracy.

[0017] To address the shortcomings of insufficient adaptability to complex environments, a dynamic weight allocation algorithm is adopted. When the global navigation satellite system signal is blocked by tree canopy, it autonomously switches to a positioning mode dominated by laser real-time positioning and map building (track drift <0.5m per hour). Combined with visual inertial odometry, continuous positioning is achieved, ensuring measurement continuity in challenging scenarios such as steep slopes and dense forests.

[0018] To address the bottleneck in data processing efficiency, an embedded real-time computing architecture is adopted, integrating point cloud denoising (statistical filtering), feature extraction random sample consistency algorithm, and parameter calculation (diameter-to-breast-length least squares fitting) processes onto the device. This enables the generation of 12 ecological parameter reports for a single sample plot within 15 minutes, which is 20 times faster than the traditional post-processing process, providing real-time data support for dynamic monitoring of forest resources. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of a handheld forest area sample point determination and mapping method provided in the embodiments of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0022] This application provides a handheld forest area sample marker mapping device, including: shell; The lidar, visual camera, and global navigation satellite system / inertial orientation and positioning navigation system integrated navigation unit are all located on the top outer part of the housing; The processing unit is located inside the casing; When the user moves with the device, the lidar collects point cloud data, and the processing unit constructs a 3D point cloud map based on the point cloud data; the visual camera collects stereo images of the surrounding environment, and the processing unit extracts feature points from the stereo images, combining the feature points and the device's inertial data to calculate the six-degree-of-freedom pose; when the Global Navigation Satellite System (GNSS) signal is available, the GNSS / Inertial Orientation and Positioning Navigation System integrated navigation unit collects GNSS positioning signals, and the processing unit combines the GNSS positioning signals, the 3D point cloud map, and the six-degree-of-freedom pose to calculate the current position; when the GNSS signal is unavailable, the GNSS / Inertial Orientation and Positioning Navigation System integrated navigation unit collects inertial data, and the processing unit combines the 3D point cloud map, the six-degree-of-freedom pose, and the inertial data to calculate the current position. After the user selects the starting point and direction of the sample plot, the processing unit analyzes the point cloud density in the 3D point cloud map and generates candidate regions. The starting point of the sample plot is located in the candidate regions. The processing unit analyzes the 3D point cloud map in the candidate regions, identifies standing trees, and uses a genetic algorithm to optimize the boundaries of the candidate regions. The processing unit calculates the parameters of the standing trees and the candidate regions after boundary optimization, and generates a sample plot mapping report based on the calculated parameters.

[0023] For example, the device of this application adopts a modular layered architecture, with a main body size of 24cm×10cm×6cm. The outer shell is made of aerospace-grade magnesium-aluminum alloy (2mm thick) and shock-absorbing rubber composite, and the total weight is controlled within 2.2kg. The top of the outer shell uses a magnesium-aluminum alloy irregular bracket to install various sensors, and the bracket also has a built-in three-point shock-absorbing pad (Shore hardness 70A) to ensure that the optical axis angle deviation is <0.2° under ±20° tilt conditions.

[0024] The sensor at the top integrates three core modules: The 16-line LiDAR (VLP-16) features a vertical field of view of ±15°, a horizontal field of view of 360°, and a maximum scanning radius of 120m. It is rigidly connected to the housing via a four-axis gimbal to compensate for ±20° jitter errors during handheld operation. The gimbal base is connected to the housing via a dovetail slide rail and is equipped with a miniature linear motor (0.5N thrust) and a Hall sensor (0.1° accuracy). Motion prediction based on inertial data (500Hz sampling rate) completes attitude adjustment within the LiDAR scanning cycle (100ms), with point cloud registration error controlled within ±1.5cm.

[0025] Global shutter binocular vision camera (Basler ace2): resolution 1280×960@30fps, baseline distance 8cm, optical axis at a 12° angle with the lidar, ensuring simultaneous capture of the trunk base (laser scanning 0.3-1.8m height range) and canopy details (stereo vision) at an operating height of 1.5m, with an overlap rate of 85% between the two synchronously acquired areas (compared to only 40% for traditional parallel layouts). The multi-frequency global navigation satellite system / inertial orientation positioning and navigation system integrated navigation unit: NovAtelPwrPak7-E1 (supporting GPS L1 / L2, GLONASS, and BeiDou tri-frequency) and TDK ICM-42688-P IMU are integrated through a metal heat dissipation substrate to achieve a tightly coupled positioning update rate of 100Hz.

[0026] Furthermore, a clamp is installed between the lidar and the vision camera to make the angle between the optical axes of the lidar and the vision camera 12°±0.5°, ensuring that the coordinate system transformation error between the two is <3mm.

[0027] The processing unit is based on the NVIDIA Jetson AGX Xavier, equipped with 32GB of LPDDR4x memory and a 1TB NVMe SSD storage unit. The housing houses a multi-sensor synchronization controller (based on the PTPv2 protocol, with time alignment accuracy of ±5μs). The housing also features a 5.5-inch sunlight-readable touchscreen (Corning Gorilla Glass, 1000nit brightness), rugged physical buttons (IP68 protection), and dual 18650 battery compartments (hot-swappable design, 10 hours of battery life). All cables are secured via internal serpentine cable management channels, and external interfaces are limited to a Type-C data port and a waterproof Micro-HDMI display port.

[0028] The process of using the device of this application for forest area sample point determination and mapping is as follows: Phase 1: Environmental Perception and Mapping Initial calibration: S101, after the device is powered on, it will automatically perform a cold start of the Global Navigation Satellite System (locking ≥4 satellites takes <30 seconds), and can obtain the initial coordinates of WGS84 or CGCS2000 according to the user settings. S102, IMU zero bias calibration (static placement for 15 seconds, acquisition of 2000 sets of data to fit the compensation curve). S103, LiDAR beam self-test (power fluctuation of each channel <1.5%, abnormal beams are automatically shielded).

[0029] Dynamic real-time positioning and map building: S104: The user's handheld device moves at a speed of 0.8m / s. The lidar generates 200,000 point cloud data per second. The processing unit constructs a three-dimensional point cloud map (resolution 0.1m) through the lidar odometry and mapping algorithm. S105 uses a vision camera to simultaneously acquire stereo images. The processing unit uses an ORB feature-based instant localization and mapping algorithm to extract 500 feature points from each stereo image and combines them with inertial data to calculate the six-degree-of-freedom pose (frequency 30Hz). S106, when the Global Navigation Satellite System (GNSS) signal is available (HDOP (Horizontal Accuracy Factor) < 1.2, or signal strength > 35 dB-Hz), a tightly coupled fusion strategy is adopted: the processing unit extracts the GNSS pseudorange and carrier phase from the GNSS positioning signal, and calculates the laser / visual odometry based on the 3D point cloud map and six-DOF pose. The GNSS pseudorange, carrier phase, and laser / visual odometry are input into the error state Kalman filter, and the global positioning result is output (major axis of the error ellipse ± ​​0.3 m). In this process, the weights of the GNSS positioning signal, the 3D point cloud map, and the six-DOF pose are 60%, 30%, and 10%, respectively.

[0030] S107: When the Global Navigation Satellite System (GNSS) lock exceeds 10 seconds (signal strength < 25 dB-Hz), a hybrid positioning mode is activated: the processing unit combines a 3D point cloud map, six-DOF pose, and inertial data to calculate the current position. In this process, the point cloud data in the 3D point cloud map provides low-frequency, high-precision pose (5Hz) through a normal distribution transform matching algorithm, while visual inertial odometry (VIO) provides high-frequency, low-precision compensation (30Hz), with a heading drift rate < 0.1° / h. Under a 30-minute GNSS lock-out environment, the positioning drift is < 0.8m (compared to > 3m drift in traditional fixed-weight schemes). In this process, the weights of the 3D point cloud map, six-DOF pose, and inertial data are 70%, 25%, and 5%, respectively. This ultimately improves the stability of the positioning result by 2.3 times.

[0031] Phase 2: Interactive Sample Marking User selection: S201: On the 3D point cloud map displayed on the touchscreen, the user can zoom to the target area with two fingers, and tap the screen to mark the starting point and direction of the sample plot (positioning accuracy ±2cm). This starting point is generally located at a corner of the target area. The processing unit will then automatically analyze the point cloud density (0.1m³ voxel grid statistics) and generate candidate areas (default 50m×50m).

[0032] Furthermore, the processing unit also predicts and adjusts the direction based on the user's operation trajectory (click frequency, drag speed). For example, if a certain area is clicked quickly and multiple times, the touch screen will automatically focus and magnify the point cloud of that candidate area.

[0033] Intelligent boundary generation: S202, Standing Tree Detection: Tree trunk point cloud clusters are identified using a density-based clustering algorithm (eps=0.5m, min_samples=3), shrub layers (height <1.3m) and fallen trees (normal vector tilt angle >45°) are removed, and finally the standing trees in the 3D point cloud map are identified; S203, Validity Verification: According to the "Technical Regulations for Continuous Forest Resource Inventory," each sample plot must contain ≥30 standing trees with a diameter at breast height (DBH) ≥5cm. If there are insufficient standing trees in the candidate area, the boundary of the candidate area will be automatically expanded along a spiral path (5m step) until the condition is met. S204, Boundary Optimization: A genetic algorithm (population size 200, 100 generations) is used to solve for the optimal boundary parameters of the candidate region. The objective function is: F = α * (1 - area error) + β * uniformity index Where α=0.6, β=0.4, and the uniformity index is calculated using the coefficient of variation of the tree spacing. Optimization time <45 seconds (Jetson AGX platform).

[0034] Specifically, the genetic algorithm includes the following process: Initial population generation: 100 candidate regions are randomly generated (rectangle side length variation range ±10%). Fitness function: F = 0.6 * (1 - area error) + 0.3 * evenness index + 0.1 * standing tree density; Genetic operations: tournament selection (scale 5), two-point crossover (probability 0.8), and Gaussian mutation (σ=0.05) were employed. Compared with traditional manual site selection, the effectiveness of the sample plots (the proportion that meets national standards) increased from 65% to 92% after adopting the genetic algorithm.

[0035] During the boundary optimization process, the processing unit will use the Monte Carlo tree search algorithm accelerated by the unified computing device architecture to pre-generate 100 candidate regions. The optimal solution will be selected through parallel computing by the graphics processing unit, which reduces the time consumption by 85% compared with the central processing unit solution.

[0036] Manual correction and confirmation: S205 allows users to drag the boundary points of the candidate area to fine-tune it, and the touch screen will display the estimated parameters in real time (such as the number of standing trees and area error after the candidate area is adjusted).

[0037] Furthermore, when the processing unit detects that a user has fine-tuned the boundary point on the same side three times consecutively, it will automatically trigger the expert rule base (which has multiple built-in forestry standards) to recommend compliant adjustment schemes (such as minimum plot spacing constraints) to the user. This can reduce the time spent on manual correction from an average of 8 minutes to 2 minutes, and reduce the incidence of violations by 75%.

[0038] S206, after the user clicks to confirm, the random sample consistency verification process is triggered: 10% of the point cloud data of the standing trees are randomly selected for cylindrical fitting. If more than 5% of the points with residuals >2cm in the cylindrical fitting are selected, a recalibration prompt will be given.

[0039] Phase 3: Extraction of Ecological Parameters Point cloud preprocessing: S301, statistical filtering to remove outliers (neighborhood radius 0.2m, standard deviation threshold 1.2). S302 uses a cloth simulation filtering algorithm to separate ground point clouds (cloth resolution 0.1m, rigidity coefficient 0.5). S303, individual trees were divided using the region growth method (seed point spacing 0.5m, curvature threshold 0.05).

[0040] Parameter calculation: S304, Diameter at breast height: A point cloud slice is taken at a height of 1.3m, and the cylinder diameter is fitted using the least squares algorithm (10 iterations, residual <1cm). This cylinder diameter is the diameter at breast height of the standing tree. S305, Tree Height: Calculate the difference between the maximum value of the standing tree on the Z-axis and the ground height (accuracy ±0.1m), and take this difference as the tree height; S306, Crown width: The length of the diagonal of the smallest bounding rectangle of the projection plane is calculated from the point cloud of the crown of the projected standing tree to the horizontal plane, and this length is taken as the crown width; S307, Canopy Closure: Laser Penetration Index (LPI) = (Number of ground points / Total number of points) × 100%, combined with the HSV color segmentation result of the stereo image (threshold H∈[100,140], S>50), it is used as the canopy closure.

[0041] Report generation: S308, the processing unit calls the Matplotlib kernel to generate a PDF format sample location mapping report, which includes: The boundary coordinates of the sample plot (UTM Zone, Mercator projection zone number, 6 decimal places). 3D point cloud bird's-eye view and side view (pseudo-color rendering); Statistical table of 12 parameters (mean, standard deviation, confidence interval); Data quality assessment matrix (completeness, consistency, and accuracy scores).

[0042] This application also provides a handheld forest area transect mapping method, such as... Figure 1 As shown, the method includes: When the device moves, it acquires point cloud data and constructs a 3D point cloud map based on the point cloud data; Acquire stereo images of the surrounding environment, extract feature points from the stereo images, and calculate the six-degree-of-freedom pose by combining the feature points and inertial data; When the Global Navigation Satellite System (GNSS) signal is available, the system collects the GNSS positioning signal and calculates the current position by combining the GNSS positioning signal, a 3D point cloud map, and a six-degree-of-freedom pose. When the GNSS signal is unavailable, the system collects inertial data and calculates the current position by combining the 3D point cloud map, a six-degree-of-freedom pose, and inertial data. Obtain the starting point and direction of the sample plot selected by the user; Analyze the point cloud density in the 3D point cloud map to generate candidate regions, with the starting point of the sample plot located within the candidate regions; Analyze the 3D point cloud map in the candidate region, identify standing trees, and use a genetic algorithm to optimize the boundary of the candidate region; The parameters of the standing trees and the candidate areas after boundary optimization are calculated, and a sample location mapping report is generated based on the calculated parameters.

[0043] This application adopts a user-guided, machine-executed interactive calibration mode, dynamically optimizing plot boundaries through a forest density model to address the applicability of traditional regular plots in irregular terrain. Simultaneously, a multi-sensor weighted adaptive algorithm was developed. When the Global Navigation Satellite System (GNSS) is available, satellite positioning is used to correct inertial navigation errors; after signal loss, it switches to real-time positioning and map-building positioning dominated by laser point cloud matching, improving positioning stability by 70% compared to single-sensor solutions. Furthermore, a lightweight embedded processing pipeline was designed, integrating point cloud denoising, feature extraction, and data fusion tasks onto the device, controlling single-frame data processing latency within 200ms, achieving truly real-time field operations.

[0044] Compared to similar products on the market, this device reduces operator training time from two weeks to one day, lowers the overall cost per measurement to 20% of traditional methods, and increases the first-pass yield rate from 68% to 98%. Furthermore, by employing parallel computing, statistical filtering, ground point cloud separation, and individual tree segmentation tasks are distributed across 512 CUDA cores. Using zero-copy memory technology, point cloud transmission latency is reduced from 15ms to 1.2ms, and the processing time for single-plot data is shortened from 56 seconds to 8 seconds. This provides a practical and feasible technical path for advancing intelligent forestry surveys.

[0045] Practical verification has shown that this application has the following significant advantages over traditional manual surveying methods: I. Improved Measurement Efficiency Traditional forest plot surveys rely on a four-person team, using compasses for orientation, measuring tapes for distance, and manual recording. A single standard plot (100m x 100m) takes an average of 120 minutes to complete, and requires repeated correction of boundary errors. In contrast, the device proposed in this application, through automatic calibration with multimodal sensors and intelligent algorithm processing, allows a single operator to complete the mapping of a plot of the same area within 15 minutes, increasing efficiency by up to eight times. In a field survey in the Wuyishan Nature Reserve in Fujian, the total time for 30 plots was reduced from 69 hours using traditional methods to 7.5 hours, while eliminating manual data transcription and post-entry steps, shortening the overall work cycle by more than tenfold.

[0046] II. Measurement Accuracy Optimization Traditional compasses can have orientation errors of up to ±8° in areas with strong magnetic interference, resulting in plot boundary rotation shifts exceeding 12 meters (based on a 100m side length). Meanwhile, measuring tapes have a ranging error rate as high as 15% in complex terrain. The device proposed in this application employs a tightly coupled positioning technology combining Global Navigation Satellite System (GNSS), laser real-time positioning and mapping, and visual inertial odometry (VIO). In open areas, its positioning accuracy reaches ±0.3m, and its heading drift rate is <0.1° / h even in GNSS lock-off conditions, ensuring that the overall boundary calibration error is controlled within ±3cm. Diameter at breast height (DBH) measurement is double-protected by laser point cloud cross-section fitting and visual texture verification, improving accuracy from ±1.2cm using the traditional measuring tape method to ±0.4cm (based on measured data from the Xishuangbanna tropical rainforest in Yunnan), meeting the stringent requirement of ≤1% error in forest parameter measurement for carbon sequestration.

[0047] III. Decreasing operating costs The comprehensive cost of traditional surveys per sample plot is approximately 380 yuan (including manpower, consumables, and travel expenses for 4 people), while the cost of UAV aerial surveys is as high as 500 yuan per sample plot (including flight permits and data processing). The device proposed in this application can complete measurements of 8 sample plots per person per day, with equipment depreciation (based on a 5-year lifespan) plus energy costs of only 58 yuan per sample plot, a reduction of 85% compared to traditional methods. Taking a provincial forest resource survey (approximately 5,000 sample plots) as an example, it can save 1.61 million yuan (1.9 million yuan for traditional methods → 290,000 yuan for this application). Furthermore, the device weighs only 2.2 kg (compared to 15 kg for traditional backpack-style LiDAR equipment), reducing the incidence of musculoskeletal injuries among workers by 78% (data from health monitoring of workers in the Yichun forest area of ​​Heilongjiang).

[0048] IV. All-terrain and all-weather operation capability Addressing the shortcomings of existing technologies in terms of poor environmental adaptability, the device in this application overcomes terrain limitations through a multi-modal sensor dynamic complementary mechanism: in steep slopes >40°, the anti-shake gimbal can control point cloud registration errors to ±2cm (±9cm without a gimbal); under dense canopies (global navigation satellite system lock-off rate >90%), laser real-time positioning and map building combined with visual inertial odometry ensures continuous positioning drift <0.8m for 6 hours; equipped with IP68 protection and a 1000nit high-brightness screen, it can operate normally at night with rainfall intensity of 50mm / h and ambient illuminance <100lux. Field tests in Liangshan Prefecture, Sichuan Province during the 2023 rainy season showed that traditional methods experienced a 42% delay rate due to weather conditions, while the device in this application maintained a 98% mission completion rate.

[0049] V. Intelligent operation and low skill threshold Traditional surveying equipment requires specialized technicians to operate (e.g., total station centering and leveling, UAV flight path planning), with training periods lasting up to two weeks. The device proposed in this application employs intent-based interactive technology: users only need short training (≤4 hours) to master the touch-based calibration process and automatically generate standardized reports conforming to the "Technical Regulations for Continuous Forest Resource Inventory." An intelligent error correction module detects violations in real time (e.g., triggering an alert when the sample plot spacing is less than 50m), increasing the data compliance rate from 65% with traditional manual recording to 98%. In a pilot project at a grassroots forestry station, a 52-year-old non-professional forest ranger independently completed sample plot surveying after 3 hours of training, achieving a 100% data compliance rate.

[0050] VI. Systematic Assurance of Data Quality Traditional manual record-keeping suffers from a data error rate exceeding 15%, and paper archives are easily damaged or lost. The device proposed in this application constructs a three-layer data quality control system: real-time detection of point cloud integrity during the acquisition phase (>20,000 valid points per second); automatic removal of outliers during the processing phase (data with a diameter at breast height residual >2cm is automatically remeasured); and generation of a blockchain fingerprint (SHA-256 hash value) during the output phase to ensure data immutability. Testing with the National Forestry and Grassland Administration's database shows that the data from this device can be directly imported into the forest resource management platform, reducing manual verification workload by 95% and meeting the Class A data standard in the "Forestry Data Quality Management Standard" (LY / T 2905-2022).

[0051] VII. Breakthroughs in Green and Low-Carbon Technologies Compared to UAV aerial surveys (1.2 kWh of power consumption and 0.75 kg of CO2 emissions per takeoff and landing), the device in this application consumes only 0.08 kWh per sample plot (10 hours of flight time / 8 sample plots), reducing the carbon footprint by 90%. The lidar uses a 905 nm wavelength (reducing power consumption by 40% compared to traditional 1550 nm equipment), employs RoHS-compliant lead-free soldering technology, and has a recyclability rate of >85% after equipment disposal. In the Anji bamboo forest carbon sequestration project in Zhejiang, measurements at 500 sample plots reduced carbon emissions by 375 kg, equivalent to the annual carbon sequestration of planting 20 fir trees, contributing to forestry surveys moving towards the goal of carbon neutrality.

[0052] Through the aforementioned technological breakthroughs, the device described in this application successfully solves the long-standing systemic problems in the field of forestry resource surveys, such as "low efficiency, poor accuracy, high cost, and strong environmental constraints," providing cost-effective technical equipment support for building digital and smart forestry. According to a third-party assessment, the widespread adoption of this technology could reduce the annual monitoring cost of my country's forest resources by 720 million yuan, reduce carbon sink measurement errors by 350 million tons of CO2 equivalent per year, and increase the efficiency of ecological value conversion by more than 20%.

[0053] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0054] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A handheld forest area sample point location and mapping device, characterized in that, include: shell; The lidar, visual camera, and global navigation satellite system / inertial orientation and positioning navigation system integrated navigation unit are all located on the outer top of the housing; The processing unit is located inside the housing; When a user moves with the device, the lidar collects point cloud data, and the processing unit constructs a three-dimensional point cloud map based on the point cloud data. The visual camera acquires stereo images of the surrounding environment, and the processing unit extracts feature points from the stereo images. It then calculates the six-degree-of-freedom pose by combining the feature points with the device's inertial data. When the Global Navigation Satellite System (GNSS) signal is available, the GNSS / Inertial Orientation and Positioning Navigation System (INS / INS) integrated navigation unit acquires GNSS positioning signals. The processing unit then calculates the current position by combining the GNSS positioning signals, the 3D point cloud map, and the six-degree-of-freedom pose. When the GNSS signal is unavailable, the INS / INS / INS integrated navigation system acquires the inertial data, and the processing unit then calculates the current position by combining the 3D point cloud map, the six-degree-of-freedom pose, and the inertial data. After the user selects the starting point and direction of the sample plot, the processing unit analyzes the point cloud density in the 3D point cloud map and generates candidate regions, with the starting point of the sample plot located in the candidate regions. The processing unit analyzes the 3D point cloud map in the candidate regions, identifies standing trees, and uses a genetic algorithm to optimize the boundaries of the candidate regions. The processing unit calculates the parameters of the standing trees and the candidate regions after boundary optimization, and generates a sample plot mapping report based on the calculated parameters.

2. The handheld forest area sample marker mapping device according to claim 1, characterized in that, When the processing unit calculates the current position by combining the Global Navigation Satellite System (GNSS) positioning signal, the 3D point cloud map, and the six-DOF pose, the weights of the GNSS positioning signal, the 3D point cloud map, and the six-DOF pose are 60%, 30%, and 10%, respectively. When the processing unit calculates the current position by combining the 3D point cloud map, the six-DOF pose, and the inertial data, the weights of the 3D point cloud map, the six-DOF pose, and the inertial data are 70%, 25%, and 5%, respectively.

3. The handheld forest area sample marker mapping device according to claim 1, characterized in that, A clamp is provided between the lidar and the vision camera, so that the angle between the optical axes of the lidar and the vision camera is 12°.

4. The handheld forest area sample marker mapping device according to claim 1, characterized in that, The housing contains a memory unit, a storage unit, and a multi-sensor synchronization controller, while the outer surface of the housing contains a touch screen, buttons, a battery compartment, a data interface, and a display interface.

5. The handheld forest area sample marker mapping device according to claim 1, characterized in that, The processing unit uses a density-based clustering algorithm to identify tree trunk point cloud clusters in the 3D point cloud map, removes shrub layers and fallen trees to determine the number of standing trees in the candidate region. If the number of standing trees is insufficient, the processing unit expands the boundary of the candidate region.

6. The handheld forest area sample marker mapping device according to claim 1, characterized in that, After the processing unit completes the boundary optimization of the candidate region, the user can fine-tune the candidate region by dragging the boundary points of the candidate region. The processing unit performs cylindrical fitting on the standing point cloud in the fine-tuned candidate region. If the cylindrical fitting result exceeds the set threshold, the processing unit will generate a recalibration prompt message.

7. The handheld forest area sample marker mapping device according to claim 1, characterized in that, Before calculating the parameters of the candidate region, the processing unit preprocesses the 3D point cloud map: Statistical filtering is used to remove outliers from the 3D point cloud map; Ground point clouds are separated using a cloth-simulated filtering algorithm; Individual trees were separated using the region growth method.

8. The handheld forest area sample marker mapping device according to claim 1, characterized in that, The parameters include the diameter at breast height (DBH), tree height, and crown width of the standing trees, as well as the canopy closure of the candidate area.

9. The handheld forest area sample marker mapping device according to claim 1, characterized in that, The sample point location survey report includes boundary coordinates, a three-dimensional point cloud bird's-eye view and side view, a parameter statistics table, and a data quality assessment matrix.

10. A method applied to the handheld forest area sample marker mapping device according to any one of claims 1-9, characterized in that, include: When the device moves, it acquires point cloud data and constructs a three-dimensional point cloud map based on the point cloud data; Acquire stereo images of the surrounding environment, extract feature points from the stereo images, and calculate the six-degree-of-freedom pose by combining the feature points with inertial data; When the Global Navigation Satellite System (GNSS) signal is available, the GNSS positioning signal is collected, and the current position is calculated by combining the GNSS positioning signal, the three-dimensional point cloud map, and the six-degree-of-freedom pose. When the GNSS signal is unavailable, the inertial data is collected, and the current position is calculated by combining the three-dimensional point cloud map, the six-degree-of-freedom pose, and the inertial data. Obtain the starting point and direction of the sample plot selected by the user; Analyze the point cloud density in the 3D point cloud map to generate candidate regions, and the starting point of the sample plot is located in the candidate regions; Analyze the 3D point cloud map in the candidate region, identify standing trees, and use a genetic algorithm to optimize the boundary of the candidate region; The parameters of the standing trees and the candidate regions after boundary optimization are calculated, and a sample location mapping report is generated based on the calculated parameters.

Images