Method for detecting health state of tree based on stress waves

Abstract

The invention relates to the technical field of nondestructive testing of materials, in particular to a method for detecting the health state of a tree based on stress waves, which comprises the following steps: acquiring geometric contour physical data of a trunk to be detected; based on the contour data and in combination with wave propagation analysis, operating an optimal sensor deployment algorithm, and calculating and determining an optimal deployment position of the intelligent sensor; stress wave excitation is carried out according to the deployment position, a stress wave response signal is collected, and then multi-dimensional waveform features are extracted; establishing a trunk section discrete model based on the geometric contour, calculating an acoustic detection path by adopting a fast marching method, and generating a geometric influence matrix; and fusing the multi-dimensional waveform features and the geometric influence matrix, reconstructing and generating an acoustic characteristic distribution diagram in the trunk to be detected through an iterative inversion algorithm, and segmenting a potential defect area based on the distribution diagram to complete the output of a health level assessment report. According to the method, through optimization deployment and multi-feature fusion inversion, the accuracy of analysis of the physical properties in the tree is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of non-destructive material testing, and in particular to a method for detecting the health status of trees based on stress waves. Background Art

[0002] Assessing the internal health of trees is crucial for public safety and ecological protection. Stress wave nondestructive testing (NDT) is the current mainstream method, identifying internal defects such as decay or hollowing by measuring changes in the speed of stress waves propagating within the tree trunk. Typical inspection systems use multiple smart sensors positioned throughout the trunk to collect stress wave propagation time data. This data is then transmitted to a computer, where software generates internal tomographic images.

[0003] However, existing methods face challenges in converting raw stress wave propagation time data into high-precision internal tomographic images. Most widely used image reconstruction algorithms rely on simplified physical models. For example, they generally assume that stress waves propagate in straight lines or employ basic interpolation methods to estimate wave velocities along non-measured paths. These simplified models become increasingly limited when faced with the inherent anisotropy and high heterogeneity of wood, as well as the complexity of internal defect morphology. This directly results in reconstructed tomographic images with low resolution and blurred defect boundaries, making it difficult to accurately restore the true contours and severity of internal defects. These images may even miss or misidentify small or early-stage lesions. Furthermore, when the number of available smart sensors is limited or their spatial layout is suboptimal, algorithm performance can degrade dramatically, compromising the reliability and stability of detection results. More importantly, existing algorithms primarily focus on the single dimension of stress wave propagation time, often overlooking other potentially rich information carried by stress waves during propagation, such as waveform amplitude attenuation, changes in frequency components, and waveform distortion, which are crucial for a more detailed characterization of internal structures.

[0004] Therefore, a method for detecting the health status of trees based on stress waves is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for detecting the health status of trees based on stress waves, aiming to improve the accuracy of tree health status detection. By obtaining the geometric contour physical data of the trunk to be tested; based on the contour data and combined with wave propagation analysis, running the optimal sensor deployment algorithm, calculating and determining the optimal deployment position of the intelligent sensor; performing stress wave excitation according to the deployment position, and collecting the stress wave response signal, and then extracting multi-dimensional waveform features; establishing a trunk cross-section discrete model based on the geometric contour, using the fast marching method to calculate the acoustic detection path and generate a geometric influence matrix; fusing the multi-dimensional waveform features and the geometric influence matrix, reconstructing the acoustic characteristic distribution map inside the trunk to be tested through an iterative inversion algorithm, and segmenting the potential defect area based on this distribution map, completing the output of the health level assessment report. The present invention improves the accuracy of the analysis of the internal physical properties of trees through optimized deployment and multi-feature fusion inversion.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for detecting the health status of trees based on stress waves, comprising:

[0008] Obtaining physical data of the geometric outline of the tree trunk to be measured;

[0009] Based on the geometric profile physical data and preset analysis conditions, combined with wave propagation analysis, an optimal sensor deployment algorithm is run to calculate the optimal deployment position of the smart sensor;

[0010] According to the optimal deployment position, stress wave excitation is sequentially performed to collect stress wave response signals; and multi-dimensional waveform features are extracted according to the stress wave response signals;

[0011] Establishing a discrete model of the trunk cross section based on the geometric profile physical data, for calculating the acoustic detection path using a fast marching method and generating a geometric influence matrix;

[0012] Combining the multi-dimensional waveform characteristics with the geometric influence matrix, an iterative inversion algorithm is used to reconstruct and generate an acoustic characteristic distribution map inside the trunk to be tested;

[0013] Based on the acoustic characteristic distribution map, potential defect areas are segmented and a health level assessment report is output.

[0014] Furthermore, the calculation process of the optimal deployment position includes:

[0015] The geometric contour physical data is used to generate a two-dimensional closed contour line through point cloud processing software;

[0016] inputting the two-dimensional closed contour and the number of smart sensors into the optimal sensor deployment algorithm;

[0017] The operation process of the optimal sensor deployment algorithm includes:

[0018] Assuming that the stress wave propagates in a straight line between the smart sensor pairs, the acoustic detection path under the current angle combination is calculated, and the total coverage length and stability index are calculated based on the acoustic detection path to obtain the fitness value;

[0019] The optimal sensor deployment algorithm performs its selection, crossover and mutation operations based on the fitness values ​​of all individuals, generates the next generation population and performs iterative search;

[0020] When the optimal sensor deployment algorithm meets the first convergence condition, the sensor angle combination is output as the optimal deployment position.

[0021] Furthermore, the process of extracting multidimensional waveform features includes:

[0022] performing frequency band filtering on the stress wave response signal;

[0023] Applying an AIC algorithm to the filtered stress wave response signal to calculate the material sound velocity propagation time;

[0024] intercepting a response signal of the stress wave response signal within a preset time window, calculating the root mean square energy of the response signal, and calculating the relative attenuation of the sound velocity in combination with the path distance;

[0025] Performing a fast Fourier transform on the response signal to obtain a frequency power spectrum, and calculating the material response center frequency;

[0026] The material sound velocity propagation time, the relative sound velocity attenuation and the material response center frequency are compiled to form the multi-dimensional waveform feature.

[0027] Furthermore, the process of calculating and generating the propagation path and geometric influence matrix includes:

[0028] Based on the geometric contour physical data, a Delaunay triangulation algorithm is used to generate a discrete model of the trunk cross section;

[0029] Setting anisotropic wave velocity on each grid cell of the trunk cross-section discrete model;

[0030] Based on the anisotropic wave velocity, the fast marching method is used to calculate the shortest propagation time field and determine the acoustic detection path between each smart sensor pair;

[0031] The travel length of each acoustic detection path in each grid unit is calculated, and the geometric influence matrix is ​​constructed and output.

[0032] Furthermore, the process of adopting the iterative inversion algorithm includes:

[0033] Initialize the wave velocity distribution of the discrete model of the trunk cross section;

[0034] Based on the geometric influence matrix and the measured propagation time extracted from the multi-dimensional waveform features, a synchronous iterative reconstruction technique is used to perform iterations; wherein, in each iteration, a predicted propagation time is forward calculated using the current wave velocity distribution and the geometric influence matrix; the difference between the predicted propagation time and the measured propagation time is calculated; and based on the difference, the wave velocity distribution is adjusted and updated;

[0035] When the difference satisfies the second convergence condition, the iteration is terminated, and the final updated wave velocity distribution is used as the acoustic characteristic distribution map.

[0036] Furthermore, the process of segmenting the potential defect area includes:

[0037] Performing preliminary segmentation on the acoustic characteristic distribution map according to a preset wave velocity threshold;

[0038] For the acoustic characteristic distribution map after preliminary segmentation, regional connectivity analysis is used to merge and identify continuous abnormal areas and mark the potential defect areas.

[0039] Furthermore, the process of outputting the health level assessment report includes:

[0040] Determining geometric and physical characteristics of the potential defect area, wherein the geometric and physical characteristics include area ratio and average wave velocity value;

[0041] According to the preset association rules between the acoustic characteristics and structural status of wood, the health level of the tree trunk to be tested is determined based on the geometric physical characteristics, and the health level assessment report is generated.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. This invention is designed at the measurement planning and data acquisition levels. On the one hand, an optimal sensor deployment algorithm based on the geometric contours of tree trunks and the simulation of acoustic wave physical propagation is introduced to make the collection of stress wave response signals more targeted, ensuring accurate and efficient data acquisition from the source. On the other hand, the collected stress wave response signals are deeply processed, not only extracting the conventional parameter of propagation time, but also mining multidimensional waveform characteristics such as the relative attenuation of sound velocity and the center frequency of material response. This constructs a physical feature set that is richer and more comprehensive than a single time parameter, laying a solid data foundation for the subsequent precise analysis of the internal physical properties and structural state of trees.

[0044] 2. This invention is designed with precision in physical modeling and image reconstruction. Based on the geometric outline of the tree trunk, a discretized model is carefully constructed, and the fast marching method is used to calculate the anisotropic acoustic detection path and geometric influence matrix. Furthermore, the multidimensional waveform features are deeply integrated with the geometric influence matrix. Using an iterative inversion algorithm, the acoustic characteristic distribution map within the tested tree trunk can be reconstructed with higher accuracy. This fusion method improves the resolution of the tomographic images and enhances sensitivity to structural anomalies within the tree, thereby more accurately revealing potential defects.

[0045] 3. This invention incorporates intelligent and quantitative design aspects into health assessment. Not only can it automatically and accurately segment potential defect areas based on the reconstructed acoustic property distribution map, but it can also further determine key geometric and physical characteristics of these areas, such as their area percentage and average wave velocity. Furthermore, using pre-defined association rules between the acoustic properties and structural state of wood, these quantitative indicators are comprehensively analyzed and determined, ultimately outputting a specific and unambiguous health assessment report. This transforms complex image interpretation and empirically based subjective judgment into standardized data analysis and rigorous rule-based reasoning, thereby improving the accuracy of the analysis of the tree's internal physical properties. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The present invention provides a flow chart of a method for detecting the health status of trees based on stress waves;

[0047] Figure 2 A schematic diagram of the process of implementing the optimal sensor deployment algorithm of the present invention;

[0048] Figure 3 The figure is a flow chart of the process of calculating and generating the geometric influence matrix according to the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] See also Figures 1 to 3 The present invention provides a method for detecting the health status of trees based on stress waves. The technical solution is as follows:

[0051] Example 1:

[0052] With the increasing awareness of ecological environmental protection, it has become a vital task to accurately and timely assess the health status of urban roadside trees, park greening trees, and ancient trees with profound historical and cultural value. However, traditional tree health detection methods mostly rely on manual experience observation, which often only stays on the surface and makes it difficult to deeply explore the internal structural conditions of the trunk. At the same time, destructive detection methods such as drilling sampling, although they can obtain some internal information, may cause secondary damage to the trees themselves, especially for those precious ancient trees, this damage is even more unbearable. Stress wave non-destructive testing technology, as an innovative means of effectively detecting internal defects of trees (such as decay, cavities, etc.), has gradually received widespread attention.

[0053] Despite this, the existing stress wave detection methods still have room for improvement in terms of the scientific nature of sensor layout, the utilization rate of characteristic information of collected signals, and the accuracy of inversion imaging. These problems limit the accuracy and reliability of the detection results to a certain extent. In order to solve the above problems, this embodiment is committed to achieving a more accurate and reliable assessment of the internal physical properties and structural status of trees through a series of measures such as optimizing data acquisition, deepening feature extraction, precise physical modeling, and intelligent image reconstruction. Figure 1 As shown, a method for detecting the health status of a tree based on stress waves includes:

[0054] The purpose of obtaining the geometric contour physical data of the tree trunk to be tested is to obtain the precise geometric shape of the tree at the detection section, providing basic support for the subsequent optimized deployment of sensors and physical modeling.

[0055] In this example, an ancient camphor tree with a diameter at breast height of approximately 80 cm was selected as the test trunk. Inspection was performed at a height of 1.5 meters above the ground. A handheld laser scanner was used to perform a 360° scan along the trunk's circumference, collecting 3D point cloud data. This data constitutes the required geometric and physical data. During the acquisition process, the integrity and accuracy of the data coverage were ensured to meet the requirements of subsequent analysis.

[0056] Based on the geometric profile physical data and preset analysis conditions, combined with wave propagation analysis, the optimal sensor deployment algorithm is run to calculate the optimal deployment position of the smart sensor to ensure the quality of subsequent stress wave data acquisition and the accuracy of inversion imaging.

[0057] In this embodiment, the preset analysis conditions include: using 12 smart sensors, expecting detection to achieve a lateral resolution of 5 cm, and the sensor layout needs to cover the entire detectable area within the cross section of the tree trunk as much as possible.

[0058] Further, Figure 2This is a flow chart of the optimal sensor deployment algorithm implemented in the present invention. Figure 2 As shown, the calculation process of the optimal deployment position includes:

[0059] First, the geometric contour physical data is used to generate a two-dimensional closed contour line through point cloud processing software;

[0060] Among them, in this embodiment, the processing process includes: importing the collected point cloud data into CloudCompare point cloud processing software, and performing the following operations in sequence: using the multi-view automatic alignment and stitching function to complete the spatial alignment of point cloud data under different scanning postures; filtering the stitched point cloud (such as removing outliers) to improve data quality; extracting the tree trunk cross-section point cloud at a height of 1.5 meters from the processed point cloud; performing contour fitting on the cross-section point cloud to generate an accurate two-dimensional closed contour line, which is stored in the form of a digital coordinate sequence (for example, angle and radius in polar coordinates, or a point set in Cartesian coordinates).

[0061] The two-dimensional closed contour and the number of smart sensors are input into the optimal sensor deployment algorithm. In this embodiment, a genetic algorithm (GA) is used as the optimal sensor deployment algorithm.

[0062] To guide the evolution of the genetic algorithm, a fitness function is set for the optimal sensor deployment algorithm. This fitness function is used to maximize the total coverage length of the acoustic detection paths between all smart sensor pairs to improve the integrity and information content of the detection area. It also minimizes a stability index to quantify the structural sensitivity and decoupling ability of the current deployment scheme during the physical inversion process.

[0063] The operation process of the optimal sensor deployment algorithm includes:

[0064] Several deployment schemes (e.g., 100 individuals) are randomly generated in the initial population. Each individual represents a possible sensor deployment method, consisting of 12 angles or position coordinates on a two-dimensional closed contour line.

[0065] For each individual in the population, the stress wave is assumed to propagate along a straight line between the smart sensor pairs (a simplified wave propagation analysis used for rapid evaluation within the genetic algorithm). Based on this straight-line propagation assumption, the simulated acoustic detection paths formed between all sensor pairs are calculated for the current individual's sensor angle combination. The total coverage length of these paths within the tree trunk cross-section is then calculated. To assess the stability of the current deployment scheme, a path geometry distribution stability index is further calculated. This index is calculated by: first, temporarily discretizing the interior of the two-dimensional closed contour of the tree trunk into a preset number of virtual cells. Then, the number of times each simulated straight path passes through these virtual cells is counted. Based on the number of times each cell is penetrated by a path, the average number of penetrations, representing the density of path coverage, and the standard deviation of the number of penetrations, representing the uniformity of path coverage, are calculated. This combined measure of path coverage density and uniformity is then derived. Second, the internal intersections between all simulated straight paths are analyzed, and the angles between the paths at these intersections are calculated. The distribution of these angles is then statistically analyzed, for example, by counting the number of intersections in different angle intervals or calculating the entropy of the angle distribution to quantify the diversity of path intersection angles. Finally, the calculated total coverage length and the path geometric distribution stability index are substituted into the fitness function through a preset weighted summation method to obtain the fitness value of the individual.

[0066] The optimal sensor deployment algorithm performs selection, crossover, and mutation operations based on the fitness of all individuals to generate the next generation of populations and conduct an iterative search. Specifically, individuals with higher fitness values ​​have a greater probability of being selected for the next generation (for example, using a roulette wheel selection method). Selected individuals are paired and exchange some of their deployment location information with a certain probability to produce new offspring individuals. This selection, crossover, and mutation operation generates a new generation of populations, and the iterative search is repeated until a termination condition is met.

[0067] When the optimal sensor deployment algorithm satisfies the first convergence condition, that is, the number of population iterations reaches a predetermined number, such as 200 times, the sensor angle combination is output as the optimal deployment position.

[0068] By accurately modeling the geometry of the tree trunk cross section and integrating acoustic wave propagation and inversion characteristics, the sensor placement is automatically optimized. This improves the integrity and representativeness of the stress wave data collected, effectively avoiding blind spots and information redundancy. Furthermore, the optimization objective encompasses inversion stability factors, helping to improve noise immunity and structural resolution in subsequent tomographic imaging.

[0069] According to the optimal deployment position, stress wave excitation is performed in sequence, and stress wave response signals are collected; and multi-dimensional waveform features are extracted according to the stress wave response signals.

[0070] Specifically, first, a smart sensor (for example, the sensor numbered S1) is selected as the current excitation point. The operator holds a standard impact hammer with a force sensor and performs a short and moderately energetic tapping action precisely near the S1 sensor and perpendicular to the surface of the tree trunk. The force sensor equipped with the impact hammer can accurately record and output the moment of the tapping (denoted as T0), which will serve as the time reference benchmark for the entire signal acquisition process. At the same time, all 12 smart sensors, including the S1 sensor as the excitation point, are started synchronously and begin to collect the response signal of the stress wave. Among them, the tapping point near the S1 sensor serves as the reference excitation source, and the remaining 11 sensors serve as receiving points, responsible for capturing and recording the response of the stress wave after propagating inside the tree trunk. After completing the signal acquisition with S1 as the excitation point, other smart sensors are selected as excitation points in turn, and the above excitation and acquisition process is repeated until the stress wave response signals covering all predetermined paths are collected.

[0071] In this example, if each deployed smart sensor acts as an excitation source, a total of 12 independent excitation events will be executed. For each excitation event, except for the excitation point itself, the remaining 11 sensors act as receivers. Therefore, the entire process ultimately accumulates 12 × 11 = 132 stress wave response signals covering different propagation paths.

[0072] Furthermore, the process of extracting multidimensional waveform features includes:

[0073] The stress wave response signal is subjected to frequency band filtering, specifically comprising applying a 4th-order Butterworth digital bandpass filter to the original stress wave response signal. Based on the empirical characteristics of stress wave propagation in ancient camphor trees, the passband range of the filter in this embodiment is set to 500 Hz to 50 kHz.

[0074] The AIC algorithm is applied to the filtered stress wave response signal to automatically identify the first arrival time of the P wave (longitudinal wave) and calculate the material sound velocity propagation time. Specifically, the AIC algorithm can accurately calibrate the inflection point where the signal emerges from the background noise, that is, the first arrival time of the P wave T by analyzing the statistical characteristics of the signal sequence, such as the change in the residual variance of the autoregressive model. i , combined with the known excitation time T0, the material sound speed propagation time T = T i -T0.

[0075] In order to quantify the energy loss of stress waves during propagation, the stress wave response signal is intercepted within a preset time window, the root mean square energy of the response signal is calculated, and the relative attenuation of the sound speed is calculated in combination with the path distance. Specifically, T iAs the starting point, intercept the response signal within a preset time window (5 milliseconds in this embodiment). Then, find the maximum absolute amplitude value within the 5 millisecond response signal and record it as V p , combined with the geometric distance L of the current signal propagation path obtained by accurate calculation of the sensor coordinates, the relative attenuation of the sound speed A is calculated. This embodiment adopts a simplified calculation method based on the peak amplitude: A = V p At the same time, the root mean square energy value of the 5 millisecond response signal can also be calculated as an auxiliary energy attenuation evaluation parameter.

[0076] For the same intercepted response signal, a fast Fourier transform is performed to obtain the frequency power spectrum, and the center frequency of the material response is calculated; the center frequency can be defined in a variety of ways, and this embodiment uses the energy-weighted average frequency of the power spectrum for calculation.

[0077] For each measurement path, the calculated material sound propagation time T, relative sound attenuation A, and material response center frequency, along with path identification information (such as the excitation and receiving point numbers), are compiled and stored as a structured data record. The data records for all paths together form a multidimensional waveform feature used for subsequent image reconstruction and health status assessment.

[0078] By processing the collected stress wave response signals, features such as propagation time, relative attenuation of acoustic velocity, and the central frequency of the material response are extracted. These features more comprehensively reflect the interaction between stress waves and the tree's internal physical properties, improving the ability to identify early lesions and subtle defects. This provides high-quality data for subsequent image reconstruction algorithms, helping to improve the accuracy and resolution of acoustic property distribution maps, thereby enabling more precise tree health assessments.

[0079] A tree trunk cross-section discrete model is established based on the geometric profile physical data, which is used to calculate the acoustic detection path using a fast marching method and generate a geometric influence matrix.

[0080] Further, Figure 3 The figure is a flow chart of the process of calculating and generating the geometric influence matrix of the present invention. Figure 3 As shown in Figure 2, the process of calculating and generating the propagation path and geometric influence matrix includes:

[0081] First, based on the geometric contour physical data, a Delaunay triangulation algorithm is used to generate a discrete model of the trunk cross section. The Delaunay triangulation algorithm ensures that the generated triangular units are as close to equilateral as possible, thereby avoiding the occurrence of overly narrow or deformed units that are disadvantageous in numerical calculations. In this embodiment, to balance calculation accuracy and subsequent calculation efficiency, the maximum side length of the generated grid units is set to no more than 2 cm. After the triangulation is completed, the structured discrete model data of the trunk cross section will be output. The model specifically includes:

[0082] Node information: records the precise spatial coordinates of each grid node (for example, the x, y coordinates in a two-dimensional Cartesian coordinate system).

[0083] Unit information: records the geometric shape of each triangular unit and the number of its constituent nodes, that is, the topological relationship of the unit.

[0084] Geometric properties: Based on the node and unit information, the geometric properties of each unit, such as area and side length, can be further calculated.

[0085] Considering that wood is a kind of natural biological material, its acoustic properties usually show significant anisotropy, that is, there is difference in the propagation speed of sound wave in different directions (for example, conforming to the wood grain direction, perpendicular to the radial direction of texture, perpendicular to the chord direction of texture).Therefore, on each grid cell of the previously generated trunk cross-section discrete model, all need set its anisotropic wave velocity parameter.In the present embodiment, according to the empirical reference value of ancient camphor tree, setting the wave velocity along trunk radial direction (being approximately equal to a principal axis of model coordinate system, for example horizontal direction) is 2.2km / s, and the wave velocity along trunk chord direction or another principal axis direction (for example vertical direction) relevant to main grain direction is 1.8km / s.

[0086] Next, based on the anisotropic wave velocity, the Anisotropic Fast Marching Method (FMM) is used to calculate the shortest propagation time field and determine the acoustic detection path between each pair of smart sensors.

[0087] FMM is an efficient numerical method used to solve the eikonal equation and calculate the shortest propagation time of the wavefront from the source point to each point in the grid. The specific calculation process is as follows:

[0088] For each smart sensor position that serves as an excitation source, it is used as the starting point for the FMM calculation. On a discrete grid, the FMM algorithm gradually expands and calculates the shortest propagation time from the excitation source to all other nodes in the grid based on the preset anisotropic wave velocity of each unit (that is, the wave velocity will change dynamically as the propagation direction changes relative to the unit's main axis direction), forming a shortest propagation time field. After obtaining the shortest propagation time field covering the entire grid, for each receiving smart sensor position with this excitation point as the source, the path is traced back from the receiving sensor position along the negative gradient direction of the time field (that is, the direction in which the time value decreases the fastest) until the excitation source position is reached. This backtracking trajectory is then determined as the acoustic detection path between this specific "excitation-receiving" smart sensor pair. This process is repeated for all valid sensor pairs to obtain a complete set of acoustic detection paths.

[0089] Finally, the travel length of each acoustic detection path in each grid unit is calculated, and the geometric influence matrix is ​​constructed and output. The specific process includes:

[0090] Initialize a geometric influence matrix of size M×N, where M is the total number of acoustic detection paths (e.g., 132) and N is the total number of grid cells in the discrete model. All elements of the matrix are initially set to 0. For each (i-th, i=1…M) calculated acoustic detection path: accurately track which grid cells the path passes through in sequence. Calculate the travel length L of the path in each (j-th, j=1…N) grid cell it passes through ij . The calculated L ij Assign to the corresponding element K of the geometric influence matrix ij If path i does not pass through cell j, then K ij Keep it at 0. After all paths are processed, the constructed geometric influence matrix is ​​output. Each row of the matrix corresponds to an acoustic detection path, each column corresponds to a grid cell, and the element value is the length of the path in the cell.

[0091] By constructing a precise discrete model of the trunk cross section and calculating an anisotropic acoustic detection path and geometric influence matrix, the accuracy of tree health analysis is improved. This model and matrix accurately reflect the propagation characteristics of stress waves in wood, providing a stable and precise mathematical foundation for subsequent image reconstruction. This helps generate more reliable internal acoustic property distribution maps, providing strong support for accurate assessment of tree health.

[0092] Combining the multi-dimensional waveform characteristics with the geometric influence matrix, an iterative inversion algorithm is used to reconstruct and generate an acoustic characteristic distribution map inside the trunk to be measured.

[0093] Furthermore, the process of adopting the iterative inversion algorithm includes:

[0094] Initialize the wave velocity distribution of the discrete model of the trunk cross section; for example, set the initial wave velocity of all grid cells to a uniform empirical value, which can be based on the average sound velocity of healthy wood of the tree being tested (for example, 2.0 km / s). Alternatively, the average value of the anisotropic wave velocity parameter can be used as the initial value.

[0095] Based on the geometric influence matrix and the measured propagation time extracted from the multidimensional waveform features, the Simultaneous Iterative Reconstruction Technique (SIRT) is used for iteration. SIRT is a robust algebraic reconstruction algorithm that gradually approaches the true solution through multiple iterations. In each iteration, the predicted propagation time is forward calculated using the current wave velocity distribution and the geometric influence matrix. The difference between the predicted propagation time and the measured propagation time is calculated. Based on the difference, the wave velocity distribution is adjusted and updated. The typical update method of SIRT is to back-project the difference based on factors such as the length of the path within the cell and the total path length of the cell, and synchronously correct the slowness of these cells.

[0096] When the difference satisfies the second convergence condition, that is, reaches a preset difference threshold, such as 1 microsecond, the iteration is terminated, and the final updated wave velocity distribution is used as the acoustic characteristic distribution map.

[0097] By adopting synchronous iterative reconstruction technology, combining the geometric influence matrix and measured propagation time, and through iterative optimization, the reconstruction accuracy and detail characterization ability of the acoustic characteristic distribution map inside the tree trunk are improved. It can better fit the actual measurement data, has good robustness to noise, and enhances the adaptability and stability of the inversion results to complex internal structures and potential measurement errors, providing a more reliable imaging basis for accurately assessing the health status of trees.

[0098] Based on the acoustic characteristic distribution map, potential defect areas are segmented and a health level assessment report is output.

[0099] Furthermore, the process of segmenting the potential defect area includes:

[0100] Performing preliminary segmentation on the acoustic characteristic distribution map according to a preset wave velocity threshold;

[0101] For the acoustic characteristic distribution map after preliminary segmentation, regional connectivity analysis is used to merge and identify continuous abnormal areas and mark the potential defect areas.

[0102] Specifically, the process of preliminary segmentation includes: healthy wood of different tree species has different wave velocity ranges. In this embodiment, for ancient camphor trees, according to empirical data, the stress wave propagation speed of healthy camphor wood is usually above 1.6km / s. Traverse each pixel or grid unit in the acoustic characteristic distribution map, and compare the wave velocity value of each unit with the preset 1.6km / s wave velocity threshold. If the wave velocity value of the unit is lower than the threshold, the unit is marked as a "suspected defect" state, for example, it is assigned a value of 1 or displayed as white in the binary image; conversely, if the wave velocity value of the unit is higher than or equal to the threshold, it is marked as a "healthy area" state, for example, it is assigned a value of 0 or displayed as black. After completing the traversal of all units, a preliminary segmented binary image is obtained.

[0103] The process of calibrating the potential defect area includes: using a standard connected component labeling algorithm, such as a scan line algorithm or a flood fill algorithm based on a 4-neighborhood or 8-neighborhood. The algorithm traverses all pixel units marked as "suspected defects" in the binary image, classifies spatially connected pixel units of the same type into an independent "connected area", and assigns a unique label to each identified connected area. Subsequently, the area of ​​each marked connected area is calculated. In order to eliminate possible artifacts or unrepresentative minor anomalies, a minimum area threshold can be set; minor areas smaller than this threshold will be regarded as artifacts and eliminated, or merged with adjacent larger defect areas when specific proximity conditions are met (such as the distance is close enough and there is a reasonable possibility of physical connection). After the above filtering and merging processing, the low wave velocity area retained is finally calibrated as the potential defect area.

[0104] Initial segmentation using a preset threshold, combined with connectivity analysis, effectively identifies potential internal defect regions, improving the repeatability and consistency of defect identification. Regional connectivity analysis, in particular, helps distinguish true, continuous defect structures from isolated noise points or reconstruction artifacts, thereby reducing the misjudgment rate in tree health assessments.

[0105] Furthermore, the process of outputting the health level assessment report includes:

[0106] Determining geometric and physical characteristics of the potential defect area, wherein the geometric and physical characteristics include area ratio and average wave velocity value;

[0107] According to the preset association rules between the acoustic characteristics and structural status of wood, the health level of the tree trunk to be tested is determined based on the geometric physical characteristics, and the health level assessment report is generated.

[0108] By measuring characteristics such as the area percentage of potential defect areas and average wave velocity, and automatically determining the grade based on pre-set rules, subjective tree health assessments are transformed into objective quantitative analyses. This improves the accuracy, consistency, and repeatability of assessments. Furthermore, the generated standardized reports provide a clear basis for decision-making in tree maintenance and management, enabling accurate assessment of the tree's internal physical properties.

[0109] The present invention firstly accurately obtains the geometric contours of the tree trunk and wave propagation analysis, uses the optimal sensor deployment algorithm to determine the layout position of the intelligent sensor, and combines it with multi-dimensional waveform feature extraction to improve the quality and information content of the original data. Secondly, a fine discrete model is established based on the geometric contour, and the fast marching method is used to calculate the acoustic detection path and generate the geometric influence matrix to provide an accurate physical model for image reconstruction. Finally, combining the multi-dimensional features and the geometric influence matrix, an iterative inversion algorithm is used to reconstruct the internal acoustic characteristic distribution map, based on which the potential defect area is segmented and a quantitative health level assessment report is output. This improves the accuracy of internal defect detection in trees, the objectivity of assessment and the level of automation, and provides reliable technical support for scientific tree maintenance and risk warning.

[0110] Example 2:

[0111] As an important part of the urban ecosystem, the health of urban roadside trees is directly related to the cityscape and public safety. Compared with ancient and famous trees, roadside trees are often numerous and have more complex growth environments. They may face a variety of unfavorable factors such as underground pipeline restrictions, narrow growth space, and human damage. Therefore, it is of great practical significance to develop a method that can quickly, accurately, and batch-assess the health of its internal structure. This example is based on the method described in Example 1 to verify the applicability and effectiveness of the method of the present invention in such scenarios.

[0112] A method for detecting the health status of trees based on stress waves, comprising:

[0113] Obtaining physical data of the geometric outline of the tree trunk to be measured;

[0114] Based on the geometric profile physical data and preset analysis conditions, combined with wave propagation analysis, an optimal sensor deployment algorithm is run to calculate the optimal deployment position of the smart sensor;

[0115] According to the optimal deployment position, stress wave excitation is sequentially performed to collect stress wave response signals; and multi-dimensional waveform features are extracted according to the stress wave response signals;

[0116] Establishing a discrete model of the trunk cross section based on the geometric profile physical data, for calculating the acoustic detection path using a fast marching method and generating a geometric influence matrix;

[0117] Combining the multi-dimensional waveform characteristics with the geometric influence matrix, an iterative inversion algorithm is used to reconstruct and generate an acoustic characteristic distribution map inside the trunk to be tested;

[0118] Based on the acoustic characteristic distribution map, potential defect areas are segmented and a health level assessment report is output.

[0119] Furthermore, the process of outputting the health level assessment report includes:

[0120] Determining geometric and physical characteristics of the potential defect area, wherein the geometric and physical characteristics include area ratio and average wave velocity value;

[0121] According to the preset association rules between the acoustic characteristics and structural status of wood, the health level of the tree trunk to be tested is determined based on the geometric physical characteristics, and the health level assessment report is generated.

[0122] Specifically, the area ratio calculation process includes: for each calibrated independent potential defect area, calculating the number of pixels it occupies on the two-dimensional tomogram. According to the actual physical area represented by a single pixel (determined by the grid unit size), the number of pixels is converted into the actual physical area of ​​the defect area (for example, in square centimeters). The physical area of ​​a single defect area or the sum of the physical areas of all defect areas is divided by the total physical area of ​​the entire trunk inspection section to obtain the defect area ratio (expressed as a percentage). This is an important indicator for evaluating the structural integrity of trees.

[0123] The average velocity calculation process involves extracting the corresponding velocity values ​​for all mesh cells within each identified potential defect region on the acoustic characteristics distribution map. The arithmetic mean of these velocity values ​​is then calculated to obtain the average velocity value for the defect region. This value reflects the density or void content within the defect region.

[0124] In this embodiment, the preset association rules between the acoustic properties of wood and the structural state include:

[0125] Level 1 (Healthy): If the total area of ​​potential defect areas is less than 5%, or the average wave speed values ​​in these areas are greater than 1.6 km / s, the tree health level is judged to be "Healthy".

[0126] Level 2 (Moderate Defects): If the total area of ​​potential defect areas is between 5% and 15%, or the average wave speed value in these areas is between 1.3km / s and 1.6km / s, the tree health level is judged to be "Moderate Defects".

[0127] Level 3 (Severe Defects): If the total area of ​​potential defect areas is greater than 15%, or the average wave speed value in these areas is less than 1.3 km / s, the tree health level is judged to be "Severe Defects".

[0128] Finally, the system integrates all analysis results and automatically generates a structured health assessment report. This report includes basic information about the tree being tested, the test date and location, the main test parameters used, an acoustic characteristic distribution map with potential defect areas marked, the geometric and physical characteristics of each defect area (including area percentage and average wave velocity), and the final health grade determined based on the aforementioned rules.

[0129] To evaluate the effectiveness of the proposed method for monitoring the health of urban roadside trees, four trees with similar diameters at breast height (DBH) (29 cm to 35 cm) were selected as samples. All samples were drilled using a micro-drill resistance meter to obtain real-world data on their internal health.

[0130] As shown in Table 1, the judgments made by the inventive method are highly consistent with the micro-drill verification results and can accurately quantify the defect area percentage. The precise quantitative assessment capabilities demonstrated by the inventive method provide stronger technical support for scientific management of tree health and risk warning.

[0131] Table 1 Results verification

[0132]

[0133] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting the health status of trees based on stress waves, characterized in that: include: Obtaining physical data of the geometric outline of the tree trunk to be measured; Based on the geometric profile physical data and preset analysis conditions, combined with wave propagation analysis, an optimal sensor deployment algorithm is run to calculate the optimal deployment position of the smart sensor; According to the optimal deployment position, stress wave excitation is sequentially performed to collect stress wave response signals; and multi-dimensional waveform features are extracted according to the stress wave response signals; Establishing a discrete model of the trunk cross section based on the geometric profile physical data, for calculating the acoustic detection path using a fast marching method and generating a geometric influence matrix; Combining the multi-dimensional waveform characteristics with the geometric influence matrix, an iterative inversion algorithm is used to reconstruct and generate an acoustic characteristic distribution map inside the trunk to be tested; Based on the acoustic characteristic distribution map, potential defect areas are segmented and a health level assessment report is output.

2. The method for detecting the health status of trees based on stress waves according to claim 1, characterized in that: The calculation process of the optimal deployment position includes: The geometric contour physical data is used to generate a two-dimensional closed contour line through point cloud processing software; inputting the two-dimensional closed contour and the number of smart sensors into the optimal sensor deployment algorithm; The operation process of the optimal sensor deployment algorithm includes: Assuming that the stress wave propagates in a straight line between the smart sensor pairs, the acoustic detection path under the current angle combination is calculated, and the total coverage length and stability index are calculated based on the acoustic detection path to obtain the fitness value; The optimal sensor deployment algorithm performs selection, crossover, and mutation operations based on the fitness values ​​of all individuals, generates the next generation population, and performs iterative search. When the optimal sensor deployment algorithm meets the first convergence condition, the sensor angle combination is output as the optimal deployment position.

3. The method for detecting tree health status based on stress waves according to claim 1, characterized in that: The process of extracting multidimensional waveform features includes: performing frequency band filtering on the stress wave response signal; Applying an AIC algorithm to the filtered stress wave response signal to calculate the material sound velocity propagation time; intercepting a response signal of the stress wave response signal within a preset time window, calculating the root mean square energy of the response signal, and calculating the relative attenuation of the sound velocity in combination with the path distance; Performing a fast Fourier transform on the response signal to obtain a frequency power spectrum, and calculating the material response center frequency; The material sound velocity propagation time, the relative sound velocity attenuation and the material response center frequency are compiled to form the multi-dimensional waveform feature.

4. The method for detecting tree health status based on stress waves according to claim 1, characterized in that: The process of calculating and generating the propagation path and geometric influence matrix includes: Based on the geometric contour physical data, a Delaunay triangulation algorithm is used to generate a discrete model of the trunk cross section; Setting anisotropic wave velocity on each grid cell of the trunk cross-section discrete model; Based on the anisotropic wave velocity, the fast marching method is used to calculate the shortest propagation time field and determine the acoustic detection path between each smart sensor pair; The travel length of each acoustic detection path in each grid unit is calculated, and the geometric influence matrix is ​​constructed and output.

5. The method for detecting tree health status based on stress waves according to claim 1, characterized in that: The process of adopting the iterative inversion algorithm includes: Initialize the wave velocity distribution of the discrete model of the trunk cross section; Based on the geometric influence matrix and the measured propagation time extracted from the multi-dimensional waveform features, a synchronous iterative reconstruction technique is used to perform iterations; wherein, in each iteration, a predicted propagation time is forward calculated using the current wave velocity distribution and the geometric influence matrix; the difference between the predicted propagation time and the measured propagation time is calculated; and based on the difference, the wave velocity distribution is adjusted and updated; When the difference satisfies the second convergence condition, the iteration is terminated, and the final updated wave velocity distribution is used as the acoustic characteristic distribution map.

6. The method for detecting tree health status based on stress waves according to claim 1, characterized in that: The process of segmenting the potential defect area includes: Performing preliminary segmentation on the acoustic characteristic distribution map according to a preset wave velocity threshold; For the acoustic characteristic distribution map after preliminary segmentation, regional connectivity analysis is used to merge and identify continuous abnormal areas and mark the potential defect areas.

7. The method for detecting tree health status based on stress waves according to claim 1, characterized in that: The process of outputting a health level assessment report includes: Determining geometric and physical characteristics of the potential defect area, wherein the geometric and physical characteristics include area ratio and average wave velocity value; According to the preset association rules between the acoustic characteristics and structural status of wood, the health level of the tree trunk to be tested is determined based on the geometric physical characteristics, and the health level assessment report is generated.

Images