Tree root system health detection method and device based on ground penetrating radar, medium and equipment

Abstract

The invention relates to the technical field of tree health detection, and particularly discloses a tree root system health detection method and device based on a ground penetrating radar, a medium and equipment, and the method comprises the steps: obtaining an original radar signal gather of a to-be-detected tree, and carrying out the synchronous preprocessing; performing spatial position and attitude correction to obtain a corrected radar signal gather; noise suppression and signal enhancement processing; using a deep learning-based feature extraction model to identify feature signals of root system and trunk defects; and generating a three-dimensional model containing the tree root system and the health condition, and generating a detection report. According to the method, damage to trees is avoided through electromagnetic wave non-contact detection, and the method is particularly suitable for detection of protection trees such as ancient trees; the multi-frequency antenna is adaptively switched to adapt to detection of root systems and tree trunks with different depths and diameters, so that the problem that a single-frequency detection range and resolution are difficult to consider at the same time is solved; through a special signal processing algorithm, a feature extraction model developed for a root system-soil-trunk heterostructure and a three-dimensional imaging algorithm, the signal analysis capability in a complex environment is improved, and accurate identification and visual presentation of root system and trunk defects are realized.

Description

Technical Field

[0001] This invention relates to the field of tree health detection technology, specifically to a method, apparatus, medium, and equipment for detecting tree root health based on ground-penetrating radar. Background Technology

[0002] Traditional tree health testing methods have long relied on drilling and visual assessment, which have insurmountable drawbacks: (1) Significantly destructive: Drilling requires drilling holes in the trunk or roots, which not only directly damages the tree's vascular tissue and destroys the bark's protective barrier, but also allows fungi, bacteria, and pests to invade through the wounds, causing the spread of diseases and potentially causing irreversible damage to ancient trees and other rare trees. (2) Seriously insufficient accuracy: Visual assessment can only infer the health status based on external characteristics such as leaf morphology and bark integrity, and cannot obtain key data such as root distribution density, depth, and internal decay and cavities in the trunk. It is easily affected by the experience of the testing personnel, resulting in large subjective errors and a high rate of missed detections. (3) Extremely low efficiency: Traditional detection of a single tree takes 1-2 hours, and the data needs to be manually recorded and organized, which is difficult to meet the timeliness requirements of detection of tens of thousands of trees in urban greening or large areas of forest. At present, ground penetrating radar (GPR) technology is gradually being used for non-destructive testing. However, although this technology has been maturely applied in fields such as geological exploration and road engineering, it has obvious adaptation defects in the field of tree detection, mainly in the following aspects: (1) Significant signal interference: Changes in soil moisture will cause fluctuations in the electromagnetic wave attenuation rate (for every 10% increase in moisture, the signal strength decreases by 15-20%). The complex branching structure of the root system (0.5-50cm in diameter) forms multi-path reflection, causing signal superposition and confusion. The difference in dielectric constant between the xylem and phloem of the trunk (about 2-4) will also interfere with the identification of internal defects. (2) Lagging data processing: Existing algorithms are mostly designed for homogeneous media (such as soil and concrete), lacking a dedicated analytical model for the heterogeneous structure of trees (root system-soil-trunk), resulting in artifacts and insufficient resolution in imaging, making it difficult to distinguish between living roots and dead roots, and between decayed and normal xylem. (3) Insufficient portability: Traditional GPR equipment weighs more than 10kg, requiring multiple people to carry it, and operation depends on professional technicians. It is difficult to operate flexibly in complex terrains in the field (such as steep slopes and bushes).

[0003] Therefore, there is an urgent need for a high-precision and convenient non-destructive testing method, device, medium, and equipment for the health of tree roots to solve the problems existing in the prior art. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide a method, device, medium and equipment for detecting tree root health based on ground penetrating radar with high accuracy and convenient detection.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A method for detecting tree root health based on ground-penetrating radar, the method comprising: S1: Obtain the original radar signal gather of the tree to be detected, and perform synchronous preprocessing on the original radar signal gather to obtain a signal-position-attitude data packet; S2: Perform spatial position and attitude correction on the signal-position-attitude data packet to obtain a corrected radar signal gather with corrected spatial position and attitude information; S3: Perform noise suppression and signal enhancement processing on the corrected radar signal gather; S4: Using a deep learning-based feature extraction model, the feature signals of root and trunk defects in the corrected radar signal gather after noise suppression and signal enhancement processing are identified, and feature data volume containing the identification results is obtained. S5: Generate a three-dimensional model containing the tree root system and health status based on the feature data, and automatically generate a detection report.

[0006] As a further improvement to the above technical solution: The original radar signal gather includes: The original time-domain reflectance signal sequence is used to represent the change of signal amplitude over time; Original latitude and longitude coordinates and elevation coordinates; Raw triaxial acceleration and triaxial angular velocity data; Timestamps are used for the synchronization and alignment of all data.

[0007] As a further improvement to the above technical solution: The synchronous preprocessing includes: Automatic gain control is applied to the original time-domain reflection signal sequence to compensate for signal attenuation caused by beam spread and medium absorption; The original latitude and longitude coordinates and elevation coordinates are calculated in real time using carrier phase difference technology to obtain a preliminary set of spatial coordinate points with centimeter-level accuracy.

[0008] As a further improvement to the above technical solution: Spatial position and attitude correction is performed on the signal-position-attitude data packet, including: After preprocessing the original triaxial acceleration data and triaxial angular velocity data, a sensor fusion algorithm based on complementary filtering is used to fuse the preprocessed triaxial acceleration data and triaxial angular velocity data to initially calculate the antenna's real-time attitude angle data and form an initial attitude matrix. The real-time attitude angle data includes pitch angle, roll angle and heading angle. The real-time attitude angle data and the data of the preliminary spatial coordinate point set are deeply fused by the extended Kalman filter algorithm to obtain the optimized trajectory and optimized attitude matrix of the ground penetrating radar. The optimized attitude matrix is ​​used to correct the receiving geometry of each radar signal in the signal-position-attitude data packet, resulting in a corrected radar signal gather with corrected spatial position and attitude information.

[0009] As a further improvement to the above technical solution: The corrected radar signal gather is subjected to noise suppression and signal enhancement processing, including: Wavelet transform denoising: Discrete wavelet transform is used, and the Db4 wavelet basis is selected to perform multi-scale decomposition on the corrected radar signal gather. The high-frequency detail coefficients after decomposition are soft-thresholded by Stein unbiased risk estimation thresholding method to remove random noise caused by soil moisture fluctuations. The processed signal is denoted as DWT_Signal[t]. Adaptive filtering: Apply an adaptive threshold filter based on statistical characteristics to DWT_Signal[t], and set the dynamic threshold Th=μ+kσ, where μ is the local mean of the signal, σ is the standard deviation, and k is an adjustable coefficient. Signals with amplitudes lower than Th are judged as environmental electromagnetic interference and are removed, resulting in the purified radar profile Filtered_RadarGram.

[0010] As a further improvement to the above technical solution: The method employs a deep learning-based feature extraction model to identify characteristic signals of root and trunk defects in the corrected radar signal gather after noise suppression and signal enhancement processing, including: The coordinates of Filtered_RadarGram and its corresponding optimized trajectory are input into a pre-trained deep learning feature extraction model, and the output is a feature data volume containing the recognition results. The feature data body containing the recognition results includes Root_Mask and Defect_Mask with classification labels and spatial coordinates; Root_Mask is the root system pixels and their predicted diameter categories; Defect_Mask is the abnormal region pixels and their types; The pre-trained deep learning feature extraction model is a convolutional neural network based on the U-Net architecture, which can perform semantic segmentation of radar images pixel by pixel.

[0011] As a further improvement to the above technical solution: Based on the feature data, a three-dimensional model containing the tree root system and health status is generated, and an inspection report is automatically generated, including: Point cloud generation and voxelization: Each identified point in Root_Mask and Defect_Mask is converted into a 3D point cloud based on the optimized trajectory coordinates; then the point cloud data is injected into a regular 3D mesh for voxelization to generate volume data containing density and attribute information; Volume rendering: The volume data is rendered using a ray casting algorithm. During the rendering process, a preset transfer function is called based on the pre-stored attributes in the volume data. Output results: The 3D imaging software ultimately generates an interactive 3D model of the tree root system and its health status, and automatically generates an inspection report.

[0012] A tree root health detection device based on ground-penetrating radar (GPR) is used to detect tree root health using the aforementioned GPR-based tree root health detection method. The tree root health detection device includes: The tunable antenna module adopts a split array design, comprising a high-frequency antenna and a low-frequency antenna. The high-frequency antenna is used to transmit a beam to detect shallow fibrous roots with a diameter ≥ 5mm within a 0-50cm range below the ground surface; the low-frequency antenna is used to transmit a beam to detect deep main roots with a diameter ≥ 5cm and the internal structure within a 50cm range of the trunk base; ultimately obtaining the original radar signal gather. The positioning and attitude unit integrates an RTK-GPS module and a six-axis IMU for spatial position and attitude correction. A radar host, which processes data from a frequency-tunable antenna module and generates a three-dimensional model containing tree root systems and health status. The human-computer interaction terminal is used to display radar waveforms, cross-sectional views and the three-dimensional model in real time, and can support operation command input, data marking and detection report generation.

[0013] The present invention also provides a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the tree root health detection method based on ground penetrating radar as described above.

[0014] The present invention also provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the tree root health detection method based on ground penetrating radar as described above.

[0015] Compared with the prior art, the advantages of the present invention are as follows: (1) A method for detecting the health of tree roots based on ground-penetrating radar according to the present invention includes acquiring the original radar signal gather of the tree to be detected, and performing synchronous preprocessing on the original radar signal gather to obtain a signal-position-attitude data package; performing spatial position and attitude correction on the signal-position-attitude data package to obtain a corrected radar signal gather with corrected spatial position and attitude information; performing noise suppression and signal enhancement processing on the corrected radar signal gather; using a feature extraction model based on deep learning to identify the feature signals of root and trunk defects in the corrected radar signal gather after noise suppression and signal enhancement processing, and obtaining a feature data body containing the identification results; generating a three-dimensional model containing the root system and health status of the tree based on the feature data body, and automatically generating a detection report. The method of the present invention uses non-contact electromagnetic wave detection, which does not require drilling or felling, avoiding physical damage to the tree's tissue structure. It is especially suitable for long-term detection of protected trees such as ancient trees and famous trees. The entire process from data acquisition to result output for a single tree is controlled within 15 minutes, which is more than 4 times more efficient than traditional methods. (2) By combining multi-frequency antennas with dynamic attitude compensation technology, the root positioning error is ≤3cm, the minimum size of internal defects in the trunk is ≥2cm, and the imaging clarity is improved by 40% compared with existing GPR tree detection equipment. Through tree-specific signal processing algorithms, feature extraction models and three-dimensional imaging algorithms are developed for the heterogeneous structure of root-soil-trunk, improving the signal analysis capability in complex environments and realizing accurate identification and visualization of root and trunk defects. The multi-frequency antenna adaptive switching technology is adapted to the detection of root and trunk structures of different depths and diameters, solving the problem that it is difficult to balance the detection range and resolution of a single frequency.

[0016] (3) In addition to root detection, the method of the present invention can simultaneously collect soil dielectric constant data (invert soil moisture), providing multi-dimensional data support for tree growth environment assessment. Attached Figure Description

[0017] Figure 1 This is a flowchart of a tree root health detection method based on ground-penetrating radar according to an embodiment of the present invention. Detailed Implementation The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, this embodiment provides a method for detecting the health of tree roots based on ground-penetrating radar. Taking the protection and detection of a 300-year-old ancient cypress as an example, the method includes: Equipment preparation: Turn on the radar host power, and the system will automatically complete the hardware self-test (takes 10 seconds). Select "Ancient Tree Detection Mode" through the interactive terminal. The equipment will automatically match the high-frequency antenna (12.6GHz) for shallow root system (0-50cm) detection and the low-frequency antenna (300MHz) for deep main root (0-2m) detection. The positioning module will switch to Beidou priority mode to improve positioning stability in complex environments.

[0019] Scanning Operation: Personnel wearing non-slip gloves slowly move the handheld device along a circular path with a 1m radius around the base of the ancient cypress trunk (speed 0.8m / s), maintaining a distance of 10±2cm between the tunable antenna module and the ground / trunk surface. The device's built-in distance sensor provides real-time alerts for distance deviations and simultaneously collects 360° scan data of the trunk (height range 0-150cm) for internal structure inspection. Electromagnetic waves are emitted at a frequency of 500MHz, and 200 sets of reflected signals are collected per second as the raw radar signal gather, simultaneously recording signal strength, propagation time, and spatial coordinate information.

[0020] Data processing: S1: The radar host acquires the original radar signal gather of the tree to be detected and performs synchronous preprocessing on the original radar signal gather to obtain a signal-position-attitude data packet. The original radar signal gather includes: Raw_Signal[t]: The original time-domain reflected signal sequence, used to represent the change of signal amplitude with time t; GPS_RAW(lon,lat,alt): Raw latitude, longitude, and elevation coordinates obtained in real time from the RTK-GPS module; IMU_RAW(acc_x,y,z,gyro_x,y,z): Raw three-axis acceleration and three-axis angular velocity data acquired from a six-axis IMU sensor; Timestamp: A precise timestamp used for the synchronization and alignment of all data.

[0021] The synchronous preprocessing includes: Automatic gain control (AGC) is applied to the original time-domain reflection signal sequence to compensate for signal attenuation caused by beam spread and medium absorption, ensuring that deep weak signals are not submerged; The original latitude and longitude coordinates and elevation coordinates are solved in real time using carrier phase difference technology to obtain a preliminary spatial coordinate point set P_initial(x,y,z) with centimeter-level accuracy; This stage outputs a preprocessed and time-synchronized signal-position-attitude data packet, with the structure {Timestamp, AGC_Signal[t], P_initial, IMU_RAW}.

[0022] S2: Perform spatial position and attitude correction on the signal-position-attitude data packet to obtain a corrected radar signal gather with corrected spatial position and attitude information, specifically including: The positioning and attitude unit records the latitude, longitude, and elevation data of the scanning path in real time, while the IMU sensor dynamically captures changes in antenna attitude. An attitude fusion algorithm corrects the angle shift caused by hand tremors and terrain undulations, ensuring that each signal corresponds to an accurate spatial position.

[0023] After preprocessing the IMU_RAW data (denoising and zero bias correction), a sensor fusion algorithm based on complementary filtering is used to fuse the preprocessed triaxial acceleration data with the triaxial angular velocity data to initially calculate the antenna's real-time attitude angles, including pitch angle θ, roll angle φ, and heading angle ψ, forming the initial attitude matrix R_imu; GPS-IMU tight coupling: The high-frequency attitude and velocity information calculated by the IMU is deeply fused with the low-frequency but absolutely accurate P_initial coordinate information provided by RTK-GPS using the Extended Kalman Filter (EKF) algorithm. The state vector X of EKF is usually defined as: X=[p x ,p y ,p z ,v x ,v y ,v z ,q0,q1,q2,q3]; Where [p] x ,p y ,p z ] represents the position, [v x ,v y ,v z [q0, q1, q2, q3] represents velocity, and [q0, q1, q2, q3] represents quaternions representing attitude. The received geometry of each radar signal is corrected using R_fused to ensure that its spatial position and incident direction are consistent with the real physical path. Through the prediction and update steps of EKF, the accumulated error of IMU and the noise of GPS are effectively estimated and compensated, and the smooth and accurate optimized trajectory P_corrected(x,y,z) and high-precision attitude matrix R_fused are output.

[0024] Finally, a corrected radar signal gather with precise spatial position and attitude information is obtained, with the structure {P_corrected,R_fused,AGC_Signal[t]}.

[0025] S3: Perform noise suppression and signal enhancement processing on the corrected radar signal gather; specifically including: Wavelet transform denoising: Discrete wavelet transform (DWT) is used. The Db4 wavelet basis is selected to perform multi-scale decomposition on the corrected radar signal gather AGC_Signal[t]. The Stein unbiased risk estimation (SURE) thresholding method is used to perform soft thresholding on the high-frequency detail coefficients after decomposition to remove random noise caused by soil moisture fluctuations. The signal-to-noise ratio is improved by 30%. The processed signal is denoted as DWT_Signal[t]. Adaptive filtering: Apply an adaptive threshold filter based on statistical characteristics to DWT_Signal[t], and set the dynamic threshold Th=μ+kσ, where μ is the local mean of the signal, σ is the standard deviation, and k is an adjustable coefficient. Signals with amplitudes lower than Th are judged as environmental electromagnetic interference and are removed, resulting in the purified radar profile Filtered_RadarGram.

[0026] S4: Using a deep learning-based feature extraction model, the feature signals of root and trunk defects in the corrected radar signal gather after noise suppression and signal enhancement processing are identified, resulting in a feature data volume containing the identification results, specifically including: The coordinates of Filtered_RadarGram and its corresponding optimized trajectory are input into a pre-trained deep learning feature extraction model, and the output is a feature data volume containing the recognition results. The feature data body containing the recognition results includes Root_Mask and Defect_Mask with classification labels and spatial coordinates; Root_Mask is the root system pixels and their predicted diameter categories (<5cm, 5-10cm, ≥10cm); Defect_Mask is the abnormal region pixels and their types (such as decay, cavities). The pre-trained deep learning feature extraction model is a convolutional neural network based on the U-Net architecture, which can perform semantic segmentation of radar images pixel by pixel.

[0027] S5: Generate a three-dimensional model containing the tree root system and health status based on the feature data, and automatically generate a detection report, specifically including: Point cloud generation and voxelization: Each identified point in Root_Mask and Defect_Mask is converted into a 3D point cloud based on the optimized trajectory coordinates; then the point cloud data is injected into a regular 3D mesh for voxelization to generate volume data containing density and attribute information. Volume rendering: Raycasting algorithm is used to render volume data. During the rendering process, the preset transfer function is called according to the pre-stored attributes in the volume data (such as root diameter and defect type). Output Results: The 3D imaging software ultimately generates an interactive 3D model of the tree root system and its health status, and automatically generates an inspection report. The report quantitatively lists key parameters such as the root distribution range, total volume, and the location and size of major defects, including: Root diameter ≥10cm -> Red; Roots 5-10cm in diameter -> yellow; Root diameter <5cm -> Green; Areas with signal strength 20% lower than normal xylem (decay) -> translucent grayish-black; Areas with signal strength higher than 80% of normal xylem (cavities / foreign objects) -> bright blue highlight.

[0028] The final output of this embodiment is as follows: The human-computer interaction terminal displays the 3D imaging results, marking the number of main roots (3), the maximum root diameter (28cm), the average distribution depth (85cm), and marking a suspected decay area (4cm in diameter) at a height of 120cm on the trunk. The data supports local storage (automatic naming format: tree species-date-latitude and longitude) and cloud synchronization, and can export a PDF detection report (including imaging images and anomaly analysis suggestions).

[0029] Secondary verification: For marked abnormal areas, a low-power mode (to reduce signal interference) is used to perform local fine scanning (resolution increased to 1cm). Combined with the tree growth history record, the degree of decay is confirmed, providing accurate data basis for formulating a rejuvenation plan.

[0030] This embodiment also provides a tree root health detection device based on ground-penetrating radar, which uses the tree root health detection method based on ground-penetrating radar as described above to detect trees. The tree root health detection device includes: The tunable antenna module, as described in this embodiment, adopts a split array design, comprising a high-frequency antenna (12.6 GHz) and a low-frequency antenna (100-500 MHz). The high-frequency antenna has a beamwidth of 30° and a resolution of 2 cm, used to transmit beams to detect shallow roots with a diameter ≥ 5 mm within a 0-50 cm range below the ground surface. The low-frequency antenna has a beamwidth of 60° and a detection depth of 2 m, used to transmit beams to detect deep main roots with a diameter ≥ 5 cm and internal structures within a 50 cm range of the tree trunk base; ultimately obtaining the original radar signal gather. The antenna module in this embodiment can also use a wideband single antenna covering 100MHz-2.6GHz to replace the split dual-band antenna. This solution reduces the mechanical structure of antenna switching and lowers equipment costs by about 20%, but the signal resolution decreases by 30% (the minimum identifiable root diameter increases to 8mm), making it suitable for large-area rapid screening scenarios with lower accuracy requirements. The tunable antenna module uses a magnetic mechanical switch (lifespan ≥1000 times) or an RF electronic switch (response time ≤10ms) to achieve rapid switching. Its shell is made of wear-resistant ABS material with an IP67 protection rating.

[0031] Through the multi-frequency antenna adaptive switching technology in this embodiment, the mechanical / electronic dual-mode switch enables rapid switching between 12.6GHz high-frequency and 100-500MHz low-frequency antennas, adapting to the detection of root systems and trunk structures of different depths and diameters, and solving the problem of difficulty in balancing the detection range and resolution of a single frequency.

[0032] The positioning and attitude unit in this embodiment integrates an RTK-GPS module and a six-axis IMU (3-axis accelerometer + 3-axis gyroscope). The RTK-GPS positioning accuracy reaches ±1cm in the plane and ±2cm in elevation, supporting BeiDou / GLONASS dual-mode positioning to ensure accurate matching of the scanning trajectory and spatial coordinates. The six-axis IMU has a sampling rate of 100Hz and can detect the antenna pitch angle, roll angle, and yaw angle in real time. It dynamically compensates for handheld shaking errors within ±5° through a Kalman filter algorithm to ensure the stability of the signal incident angle and ultimately achieve spatial position and attitude correction. The dynamic attitude intelligent compensation system integrates RTK-GPS and six-axis IMU data and corrects the antenna angle offset (≤5°) in real time through a Kalman filter algorithm to ensure the spatial consistency of signal acquisition in complex operating environments and improve imaging accuracy.

[0033] If the positioning and attitude unit of this embodiment is in an area where RTK-GPS signals are limited, such as dense forests or canyons, a 200m range laser rangefinder can be used to assist in positioning. A local coordinate system can be constructed by pre-marked points, reducing the positioning accuracy to ±5cm, but the equipment cost is reduced by 40%, and it is not affected by satellite signals.

[0034] The radar host processes data from the frequency-tunable antenna module and generates a 3D model including the tree root system and its health status. The radar host adopts a modular design, with its core control module based on an ARM Cortex-A72 processor with a main frequency of 2.0GHz, supporting real-time signal processing and multi-device collaborative control. Its data storage unit is equipped with a 128GB industrial-grade SSD, supporting continuous storage of 100 hours of raw data and imaging results. The power supply system uses a removable 10Ah lithium battery, providing 8 hours of battery life, supporting fast charging (fully charged in 2 hours), and featuring a low battery warning function. The host has built-in Wi-Fi 6 and Bluetooth 5.0 modules, enabling real-time data transmission to a human-machine interface terminal or synchronization to a cloud platform.

[0035] If it is necessary to reduce the computing power requirements of the terminal, an edge computing architecture can be adopted. The host only performs signal preprocessing, and the raw data is uploaded to the cloud server for in-depth analysis. This solution reduces the size and weight of the terminal by 15%, but it depends on a stable network environment, and the real-time performance drops to the minute level.

[0036] The human-machine interface terminal is equipped with a 7-inch high-definition touchscreen (1920×1080 resolution), supports glove operation and multi-touch, and has a built-in industrial-grade heat dissipation system (operating temperature -20℃ to 50℃). The terminal comes pre-installed with 3D imaging software, which can display radar waveforms, cross-sectional views and 3D models in real time, and supports operation command input (such as antenna switching, parameter adjustment), data marking and test report generation.

[0037] This embodiment features an integrated portable architecture design: the radar host, dual-band antenna, positioning unit, and interactive terminal are integrated into a lightweight waterproof housing. Optimized power management and heat dissipation design enable single-person portable operation and long-term field work, breaking through the scenario limitations of traditional GPR equipment. The portable design adapts to complex environments. The integrated design weighs ≤5kg and measures 30×20×15cm. Equipped with a waterproof shoulder strap and IP65 protective shell, it can operate stably in rainy, muddy, and steep slope environments. The operation process is simplified to three steps: "power on - scan - view results," and ordinary maintenance personnel can operate independently after one hour of training.

[0038] This embodiment also includes a readable storage medium storing a computer program that, when executed by a processor, implements the tree root health detection method based on ground-penetrating radar as described above.

[0039] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program segments capable of performing a specific function, and these instruction segments describe the execution process of the computer program in the electronic device.

[0040] This embodiment also provides an electronic device, including: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the tree root health detection method based on ground penetrating radar as described above.

[0041] The electronic device can be a mobile phone, desktop computer, laptop, handheld computer, cloud server, or other computing device. The electronic device may include, but is not limited to, processors and memory. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0042] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. For those skilled in the art, improvements and modifications obtained without departing from the inventive concept should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting tree root health based on ground penetrating radar, characterized in that, The method comprises: S1: obtaining an original radar signal channel set of a tree to be detected, and performing synchronization preprocessing on the original radar signal channel set to obtain a signal-position-pose data packet; S2: performing spatial position and pose correction on the signal-position-pose data packet to obtain a corrected radar signal channel set with corrected spatial position and pose information; S3: performing noise suppression and signal enhancement processing on the corrected radar signal channel set; S4: using a feature extraction model based on deep learning to identify the feature signals of root systems and stem defects of the corrected radar signal channel set after noise suppression and signal enhancement processing, to obtain a feature data body containing identification results; S5: generating a three-dimensional model containing the root systems and health conditions of the tree according to the feature data body, and automatically generating a detection report. 2.The ground penetrating radar-based tree root health detection method of claim 1, wherein, The original radar signal channel set comprises: an original time-domain reflection signal sequence for representing the change of signal amplitude with time; original latitude and longitude coordinates and elevation coordinates; original three-axis acceleration data and three-axis angular velocity data; a timestamp for synchronization alignment of all data. 3.The ground penetrating radar-based tree root health detection method of claim 2, wherein, The synchronization preprocessing comprises: applying automatic gain control to the original time-domain reflection signal sequence for compensating for signal attenuation caused by beam spreading and medium absorption; real-time solving the original latitude and longitude coordinates and elevation coordinates through carrier phase difference technology to obtain a preliminary spatial coordinate point set with centimeter-level precision. 4.The ground penetrating radar-based tree root health detection method of claim 3, wherein, The spatial position and pose correction on the signal-position-pose data packet comprises: after preprocessing the original three-axis acceleration data and three-axis angular velocity data, using a sensor fusion algorithm based on complementary filtering to fuse the preprocessed three-axis acceleration data and three-axis angular velocity data, preliminarily solving real-time attitude angle data of the antenna, and forming an initial attitude matrix, wherein the real-time attitude angle data comprises a pitch angle, a roll angle and a heading angle; deeply fusing the real-time attitude angle data and the data of the preliminary spatial coordinate point set through an extended Kalman filtering algorithm to obtain an optimized trajectory of the ground penetrating radar and an optimized attitude matrix; using the optimized attitude matrix to correct the receiving geometry of each radar signal in the signal-position-pose data packet to obtain the corrected radar signal channel set with corrected spatial position and pose information. 5.The ground penetrating radar-based tree root health detection method of claim 4, wherein, The noise suppression and signal enhancement processing on the corrected radar signal channel set comprises: wavelet transform denoising: using discrete wavelet transform, selecting Db4 wavelet basis to perform multi-scale decomposition on the corrected radar signal channel set, and using Stein unbiased risk estimation threshold method to perform soft threshold processing on the decomposed high-frequency detail coefficients, for removing random noise caused by soil humidity fluctuation, and the processed signal is denoted as DWT_Signal[t]; Adaptive filtering: Apply adaptive threshold filtering based on statistical characteristics to DWT_Signal[t], set dynamic threshold Th = μ + kσ, where μ is the local mean of the signal, σ is the standard deviation, and k is the adjustable coefficient. Signals with amplitudes below Th are judged as environmental electromagnetic interference and are removed, obtaining the purified radar profile Filtered_RadarGram. 6.The ground penetrating radar-based tree root health detection method of claim 5, wherein, The feature extraction model based on deep learning is used to identify the feature signals of the root system and trunk defects of the corrected radar signal gather after noise suppression and signal enhancement processing, including: The coordinates of Filtered_RadarGram and its corresponding optimized trajectory are input into the pre-trained deep learning feature extraction model, and the feature data body containing the identification results is output; The feature data body containing the identification results includes Root_Mask and Defect_Mask with classification labels and spatial coordinates; Root_Mask is the root pixel and its predicted diameter category; Defect_Mask is the abnormal area pixel and its type; The pre-trained deep learning feature extraction model is a convolutional neural network based on U-Net architecture, which can perform semantic segmentation on radar images pixel by pixel. 7.The ground penetrating radar-based tree root health detection method of claim 6, wherein, According to the feature data body, a three-dimensional model containing tree root system and health status is generated, and a detection report is automatically generated, including: Point cloud generation and voxelization: Each identified point in Root_Mask and Defect_Mask is converted into a three-dimensional point cloud according to the optimized trajectory coordinates; then the point cloud data is injected into a regular three-dimensional grid for voxelization, generating volume data containing density and attribute information; Volume rendering: Use ray casting algorithm to render the volume data, and call the preset transfer function according to the pre-stored attributes in the volume data during rendering; Result output: The three-dimensional imaging software finally generates an interactive three-dimensional model of tree root system and health status, and automatically generates a detection report. 8.A ground penetrating radar-based tree root health detection device, characterized by, The tree root health detection method based on ground penetrating radar of any one of claims 1-7 is used to detect trees, and the tree root health detection device includes: Adjustable frequency antenna module, the adjustable frequency antenna module adopts a split array design, including high frequency antenna and low frequency antenna; the high frequency antenna is used to emit beam to detect shallow root in the range of 0-50 cm below the ground and diameter ≥5 mm; the low frequency antenna is used to emit beam to detect deep main root with diameter ≥5 cm and internal structure within 50 cm range of trunk base; and finally obtain the original radar signal gather; Positioning and attitude unit, the positioning and attitude unit integrates RTK-GPS module and six-axis IMU, which is used for spatial position and attitude correction; Radar host, the radar host is used to process data from the adjustable frequency antenna module, and generate a three-dimensional model containing tree root system and health status; Human-computer interaction terminal, the human-computer interaction terminal is used to display radar waveform graph, profile graph and three-dimensional model in real time, and can support operation instruction input, data labeling and detection report generation.

9. A readable storage medium, characterized by, A computer program product, comprising a computer readable medium having stored thereon the computer program, the computer program comprising instructions which, when executed by a processor, cause the processor to carry out the ground penetrating radar based tree root health detection method according to any one of claims 1-7.

10. An electronic device, comprising: A computer program product, comprising a computer readable medium having stored thereon the computer program, the computer program comprising instructions which, when executed by a processor, cause the processor to carry out the ground penetrating radar based tree root health detection method according to any one of claims 1-7.

Images