An intelligent method for calculating the volume of artificial forest based on laser radar point cloud

Abstract

This invention relates to the field of biomass measurement technology, specifically to an intelligent method for measuring the stock volume of plantations based on lidar point clouds. The method includes: acquiring isolated non-stand structure noise points; clustering the data points according to the distances between data points in the dataset to be processed to obtain data point clusters; obtaining cluster noise characterization values ​​for each data point cluster based on the intra-cluster nearest neighbor distances and the results of region growing for each data point cluster; filtering out non-stand structure noise clusters based on the cluster noise characterization values; removing points belonging to isolated non-stand structure noise points and non-stand structure noise clusters from the original digital surface model and original digital terrain model to obtain a target digital surface model and a target digital terrain model; and obtaining the stock volume of the plantation to be measured based on the target digital surface model and target digital terrain model. Furthermore, this invention can improve the accuracy of plantation stock volume measurement.

Description

Technical Field

[0001] This invention relates to the field of biomass measurement technology, specifically to an intelligent method for measuring the stock volume of plantations based on lidar point clouds. Background Technology

[0002] Currently, in order to promote forestry resource management and ensure ecological security and sustainable development, it is usually necessary to calculate the stock volume of planted forests. Moreover, with the development of lidar technology, the current method of calculating planted forests is to use drones equipped with lidar to collect three-dimensional point clouds of planted forests. The calculation coverage efficiency of drones equipped with lidar is high and can greatly shorten the calculation cycle.

[0003] However, when estimating the volume of plantations based on 3D point clouds collected by lidar for plantations, environmental interference (such as rainfall and fog) easily leads to a large amount of non-stand structure noise in the point cloud. The presence of non-stand structure noise will disrupt the accurate extraction of core parameters such as tree height and canopy structure, which will cause the volume measurement results to deviate significantly from the true value. In other words, when there is non-stand structure noise in the DSM (Digital Surface Model) and DTM (Digital Terrain Model), it will cause the CHM (Canopy Height Model) to deviate from the true value. Since CHM is the key to analyzing plantation biomass, deviations in volume measurement results will threaten the sustainable development of forestry and ecological security. Therefore, how to eliminate the interference of environmental noise on the accuracy of plantation biomass measurement and improve the accuracy of plantation volume measurement has become an urgent problem to be solved. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides an intelligent method for calculating the stock volume of planted forests based on lidar point clouds. The specific technical solution adopted is as follows: One embodiment of the present invention provides an intelligent method for calculating the stock volume of planted forests based on lidar point clouds, comprising the following steps: Obtain the set of original model data points corresponding to the artificial forest to be measured. The set of original model data points consists of data points from the original digital surface model and the original digital terrain model. Based on the number of points within a preset neighborhood radius of each data point in the original model data point set and the distance between each data point and each point within the preset neighborhood radius, the canopy confidence feature value of each data point is obtained. Isolated non-stand structure noise points are screened out based on the canopy confidence feature value. Isolated non-stand structure noise points in the original model data point set are removed to obtain a set to be processed. The data points are clustered according to the distance between the data points in the set to be processed to obtain each data point cluster. Based on the nearest neighbor distance within each data point cluster and the results of region growing for each data point cluster, the cluster noise characterization value of each data point cluster is obtained. Non-stand structure noise clusters are screened out based on the cluster noise characterization value. Points belonging to isolated non-stand structure noise points and non-stand structure noise clusters in the original digital surface model and original digital terrain model are removed to obtain the target digital surface model and target digital terrain model. The stock volume of the plantation to be measured is obtained based on the target digital surface model and target digital terrain model.

[0005] Beneficial effects: This invention first obtains the original model data point set corresponding to the plantation to be measured; then, based on the number of points within a preset neighborhood radius of each data point in the original model data point set and the distance between each data point and each point within the preset neighborhood radius, the canopy confidence feature value of each data point is obtained. Based on the canopy confidence feature value, isolated non-stand structure noise points are screened out. The isolated non-stand structure noise points in the original model data point set are removed to obtain the set to be processed. The data points are clustered according to the distance between the data points in the set to be processed to obtain each data point cluster. Based on the nearest neighbor distance within each data point cluster and the results of regional growth of each data point cluster, the cluster noise characterization value of each data point cluster is obtained. Non-stand structure noise clusters are screened according to the cluster noise characterization value. Finally, points belonging to isolated non-stand structure noise points and non-stand structure noise clusters in the original digital surface model and the original digital terrain model are removed to obtain the target digital surface model and the target digital terrain model. The stock volume of the plantation to be measured is obtained based on the target digital surface model and the target digital terrain model. Furthermore, this invention identifies isolated non-stand structural noise points and non-stand structural noise clusters based on canopy confidence feature values ​​and cluster noise point characterization values. It also determines the plantation stock volume to be measured based on the target digital surface model and target digital terrain model after removing isolated non-stand structural noise points and non-stand structural noise clusters. This can reduce or eliminate the interference of environmental disturbances on the accuracy of plantation biomass measurement, thereby improving the accuracy of plantation stock volume measurement. Attached Figure Description

[0006] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0007] Figure 1 This is a flowchart of an intelligent method for calculating the stock volume of planted forests based on lidar point clouds, according to the present invention. Detailed Implementation

[0008] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the protection scope of the embodiments of the present invention.

[0009] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art.

[0010] This embodiment provides an intelligent method for calculating the stock volume of planted forests based on lidar point clouds, detailed as follows: like Figure 1 As shown, this intelligent method for calculating the stock volume of planted forests based on lidar point clouds includes the following steps: Step S001: Obtain the original model data point set corresponding to the artificial forest to be measured. The original model data point set consists of data points from the original digital surface model and the original digital terrain model.

[0011] When collecting 3D point clouds of plantations using lidar, rain or fog can cause water droplets and moisture in the air to interfere with the laser signals emitted by the lidar on the drone. This results in a large number of false points (i.e., non-stand structure noise points) in the collected point cloud data, which do not reflect the true vegetation structure. The presence of this non-stand structure noise disrupts the accurate extraction of core parameters such as tree height and canopy structure. Consequently, the generated DSM (Digital Surface Model) and DTM (Digital Terrain Model) data from the plantation's 3D point cloud contain non-stand structure noise. This noise causes the calculated CHM (Canopy Height Model) values ​​from the DSM and DTM to deviate from the true values, leading to errors in the base... Since the plantation stock volume calculated by CHM deviates from the true value, CHM is the key to estimating or calculating the plantation stock volume. In order to ensure the accuracy of the plantation stock volume calculation, this embodiment will eliminate or reduce the impact of noise on the accuracy of the plantation stock volume calculation. That is, to eliminate as much non-forest structure noise as possible in DSM (Digital Surface Model) and DTM (Digital Terrain Model), the plantation stock volume will be calculated based on the CHM obtained from the denoised DSM and DTM to improve the accuracy of the plantation stock volume calculation. In other words, in order to ensure the accuracy of the plantation stock volume calculation, this embodiment will subsequently denoise the DSM and DTM generated based on the 3D point cloud of the plantation.

[0012] For ease of understanding, this embodiment will subsequently describe the process of calculating the volume of any plantation forest that requires volume measurement, denoted as the plantation forest to be measured. Then, a UAV equipped with a lidar system is used to collect a 3D point cloud of the plantation forest to be measured. This collected 3D point cloud is denoted as the original 3D point cloud of the plantation forest to be measured. The process of acquiring the 3D point cloud of the plantation forest using a UAV equipped with lidar is well-known. The X, Y, and Z coordinates of the data points in the 3D point cloud represent the eastward position, northward position, and absolute elevation of the point, respectively. The original 3D point cloud of the plantation forest to be measured is then processed to obtain a Digital Surface Model (DSM) and a Digital Terrain Model (DTM), denoted as the original digital surface model and the original digital terrain model, respectively. The coordinates of each point in the digital surface model (DSM) and the digital terrain model (DTM) have the same physical meaning as the points in the original 3D point cloud. The DTM is a skeleton model of the ground, representing the area to be measured. The continuous elevation distribution of the exposed surface of the plantation area is represented by the Digital Surface Model (DSM), which characterizes the highest continuous elevation of the canopy layer on the surface of the plantation area to be tested. In this embodiment, the Digital Surface Model (DSM) corresponds to the ground surface in the actual surveying process or the ground surface of the plantation area to be tested, and the Digital Terrain Model (DTM) corresponds to the canopy layer of the plantation in the actual surveying process or the canopy layer of the plantation area to be tested. The original Digital Surface Model and the original Digital Terrain Model are obtained by directly processing the three-dimensional point cloud of the plantation to be tested using lidar point cloud data processing and analysis software, which is a known technology. The Digital Surface Model (DSM) and the Digital Terrain Model (DTM) are continuous and structured surface models generated by classifying, interpolating, and meshing the three-dimensional point cloud of the plantation to be tested.

[0013] Next, all data points in the original digital surface model and the original digital terrain model are acquired and set as the original model data point set corresponding to the artificial forest to be measured. Alternatively, a three-dimensional space is constructed first. The meanings of the X-axis, Y-axis, and Z-axis in the three-dimensional space are consistent with the meanings of the X-coordinate, Y-coordinate, and Z-coordinate of the data points in the three-dimensional point cloud in this embodiment. All points in the original digital surface model and all points in the original digital terrain model are mapped to the three-dimensional space. The set of all data points in the three-dimensional space after mapping is set as the original model data point set corresponding to the artificial forest to be measured. Furthermore, the subsequent neighborhood analysis and distance measurement of the data points are all performed in a unified three-dimensional space.

[0014] Therefore, this embodiment can obtain the original model data point set corresponding to the artificial forest to be measured through the above process. Subsequently, the original model data point set will be analyzed and noise points in the original model data point set will be identified.

[0015] Step S002: Based on the number of points within a preset neighborhood radius of each data point in the original model data point set and the distance between each data point and each point within the preset neighborhood radius, the canopy confidence feature value of each data point is obtained. Isolated non-stand structure noise points are screened out based on the canopy confidence feature value. Isolated non-stand structure noise points in the original model data point set are removed to obtain a set to be processed. The data points are clustered according to the distance between the data points in the set to be processed to obtain each data point cluster. Based on the nearest neighbor distance within each data point cluster and the results of region growing for each data point cluster, the cluster noise characterization value of each data point cluster is obtained. Non-stand structure noise clusters are screened out based on the cluster noise characterization value.

[0016] During lidar transmission, weather conditions, such as the refraction of radar signals by water droplets, can generate outlier noise points, also known as isolated non-forest stand noise points. Compared to data points normally located within the canopy, these isolated non-forest stand noise points are farther from other data points and have fewer neighboring data points. Based on the above description, the number of points within a preset neighborhood radius of a data point in three-dimensional space and the distance between the data point and each point within that radius can reflect whether the corresponding data point is an isolated non-forest stand. The possibility of a structural noise point being a data point within the forest canopy is considered. Therefore, in this embodiment, the canopy confidence feature value of each data point in the original model data point set will be obtained based on the number of points within a preset neighborhood radius of each data point in the three-dimensional space and the distance between each point and the preset neighborhood radius. The canopy confidence feature value can reflect the possibility that the corresponding data point is an isolated non-stand structural noise point or a data point within the forest canopy. The specific calculation process of the canopy confidence feature value of each data point in the original model data point set is as follows: First, the preset neighborhood radius is determined. In this embodiment, the specific process for determining the preset neighborhood radius is as follows: obtain the set nearest neighbor distance corresponding to each data point in the original model data point set in three-dimensional space, and select the largest set nearest neighbor distance as the preset neighborhood radius from all the set nearest neighbor distances corresponding to all data points in the original model data point set. The set nearest neighbor distance corresponding to any data point in the original model data point set is the minimum Euclidean distance among the Euclidean distances between that data point and all other data points in the original model data point set except for that data point. As another implementation method, the implementer can also set the size of the preset neighborhood radius according to the actual scenario.

[0017] Next, in 3D space, obtain the set of data points within the preset neighborhood radius of each data point in the original model data point set. Based on the neighborhood set of the corresponding data point, if the preset neighborhood radius is R0, and only data points a1, a2, and a3 in the original model data point set are located within a circle centered on data point a0 and with radius R0, then the set of data points a1, a2, and a3 is the neighborhood set of data point a0. Then, count the number of data points in the neighborhood set of each data point in the original model data point set, and record this as the neighborhood number of the corresponding data point. Finally, perform maximum and minimum value normalization on the neighborhood number of each data point. First, the normalization result is recorded as the neighborhood count measure of the corresponding data point. Next, the mean Euclidean distance between each data point in the original model data point set and each data point in its neighborhood set is calculated, and a negative correlation mapping is applied to the mean. This mapping result is recorded as the neighborhood feature value of the corresponding data point. Then, the neighborhood feature value of the data point is normalized to its maximum and minimum values, and the result is recorded as the distance representation value of the corresponding data point. Here, the negative correlation mapping to the mean refers to taking the reciprocal of the mean. Finally, the neighborhood count measure of each data point is added to the distance representation value of the corresponding data point, and this result is recorded as the canopy confidence feature value of the corresponding data point. The specific calculation process for the canopy confidence feature value of any data point q in the original model data point set is as follows:

[0018] in, Let q be the tree canopy confidence feature value. Let q be the number of neighborhood points of data point q. This is the minimum number of neighborhood points among all data points in the original model data point set. This is the maximum value among the number of neighborhood points of all data points in the original model data point set. Let be the neighborhood feature values ​​of data point q. Let be the Euclidean distance between data point q and the i-th data point in the neighborhood set of data point q. This is the minimum value among the neighborhood feature values ​​of all data points in the original model data point set. It is the maximum value among the neighborhood feature values ​​of all data points in the original model data point set. To The formula for performing maximum and minimum value normalization, Similarly; The larger the value, the more data points are located in the neighborhood of data point q in the original model data point set, which in turn indicates that data point q is more likely to be a data point inside the tree canopy. The larger the value, the closer the distance between the data point q and the points in its neighborhood, thus indicating a greater likelihood that the data point q is located within the forest canopy; because The greater the sum When it is larger, The larger, therefore The larger the value of q, the greater the probability that the data point q is located inside the tree canopy, and vice versa. The smaller the value, the less likely the data point q is to be a data point inside the tree canopy, and the greater the likelihood that it is an isolated noise point outside the forest stand structure.

[0019] Since the canopy confidence feature value can reflect the probability that the corresponding data point is an isolated non-stand structure noise point, this embodiment, after obtaining the canopy confidence feature value of the data point, filters out isolated non-stand structure noise points in the original model data point set based on the canopy confidence feature value. The specific process of filtering isolated non-stand structure noise points in the original model data point set based on the canopy confidence feature value is as follows: For any data point in the original model data point set, it is determined whether the tree canopy confidence feature value of the data point is less than the preset isolated noise threshold. If it is less than the threshold, it indicates that the data point is isolated non-stand structure noise, and the data point is determined to be an isolated non-stand structure noise point. If it is not less than the threshold, it indicates that the data point is not isolated non-stand structure noise, and it may be a data point inside the tree canopy. In specific applications, the implementer needs to set the preset isolated noise threshold according to the actual situation such as the range of tree canopy confidence feature values. For example, in this embodiment, the tree canopy confidence feature values ​​of data points inside the tree canopy and isolated non-stand structure noise points can usually be divided into the two ends of the tree canopy confidence feature value range through the above process. Therefore, in order to better identify isolated non-stand structure noise, this embodiment can use the median value of the tree canopy confidence feature value range, 1, as the preset isolated noise threshold.

[0020] However, noise points caused by meteorological interference can also form clusters. For example, when clouds and water vapor have a strong impact on radar and laser systems, their concentration often leads to localized patches or strip-shaped clusters of cloud and fog. This causes the refracted noise points generated by interference in these areas to form a relatively clustered data point cluster. Since canopy confidence features cannot identify and filter out clustered non-stand structure noise points, this embodiment, after identifying isolated non-stand structure noise points, needs to re-analyze the data point set after removing isolated non-stand structure noise points to identify clustered non-stand structure noise points. The specific process is as follows: First, isolated non-forest stand noise points are removed from the original model data point set, and the set of remaining points is denoted as the unprocessed set. Then, based on the distances between data points in the unprocessed set, K-means clustering is performed on all data points in the unprocessed set, and the resulting clusters are denoted as data point clusters. The distance between data points during clustering is the Euclidean distance, and the number of cluster centers is determined by the elbow method. The process of clustering using the K-means clustering algorithm with known distances and the process of determining the number of cluster centers based on the elbow method are well-known techniques. Finally, based on the data point distribution characteristics and region growth characteristics of each data point cluster, the cluster noise characterization value of the data point cluster is obtained. The noise level is primarily characterized by the intra-cluster nearest neighbor distance of data points within a data point cluster. Region growth characteristics are mainly characterized by the results of region growth of data point clusters. Therefore, in this embodiment, the cluster noise characterization value of each data point cluster is obtained based on the intra-cluster nearest neighbor distance of data points within each data point cluster and the results of region growth of each data point cluster. The cluster noise characterization value reflects the probability that the corresponding data point cluster is a non-forest stand structure noise cluster. Furthermore, for ease of understanding, this embodiment will describe the process of obtaining the cluster noise characterization value of any data point cluster A as an example. The specific process of obtaining the cluster noise characterization value of data point cluster A based on the intra-cluster nearest neighbor distance of data points in data point cluster A and the results of region growth of data point cluster A is as follows: First, based on the mean of the intra-cluster nearest neighbor distances of data points in data point cluster A and the mean of the absolute values ​​of the differences between the intra-cluster nearest neighbor distances of data points in data point cluster A and the intra-cluster nearest neighbor distances of other data points in the cluster, the compact uniformity characterization value of data point cluster A is obtained. The compact uniformity characterization value represents the distribution characteristics of data points in data point cluster A. The compact uniformity characterization value can reflect the probability that data point cluster A is a non-stand structure noise cluster, and it is also a key parameter reflecting whether data point cluster A is a non-stand structure noise cluster. The specific process of obtaining the compact uniformity characterization value of data point cluster A is as follows: Calculate the mean of the intra-cluster nearest neighbor distances for all data points in data cluster A, and denot it as the first intra-cluster distribution characteristic value. The intra-cluster nearest neighbor distance of any data point in any data cluster is the minimum Euclidean distance between that data point and every other data point in that cluster, or in other words, the intra-cluster nearest neighbor distance of any data point in any data cluster is the Euclidean distance between that data point and the nearest point in that cluster. Calculate the mean of the absolute values ​​of the differences between the intra-cluster nearest neighbor distances of each data point in data cluster A and the intra-cluster nearest neighbor distances of all other data points in data cluster A (excluding the corresponding data point), and denot it as the mean of the intra-cluster distances of the corresponding data point. That is, the mean of the intra-cluster distances of the j-th data point in data cluster A is the mean of the absolute values ​​of the differences between the j-th data point and the intra-cluster nearest neighbor distances of all other data points in data cluster A. The expression for the mean of the intra-cluster distances of the j-th data point is: ,in, Let be the number of data points in the set consisting of the remaining data points within data point cluster A, excluding the j-th data point. Let j be the intra-cluster nearest neighbor distance of the j-th data point. Let be the intra-cluster nearest neighbor distance of the r-th data point in the set of data points remaining in data cluster A (excluding the j-th data point). Calculate the mean of the intra-cluster distances of all data points in data cluster A, and denote it as the second intra-cluster distribution characteristic value. Calculate the normalized result of adding the first and second intra-cluster distribution characteristic values, and denote it as the compact uniform representation value of data cluster A. Here, the normalization function Norm() is used for normalization. The expression for the compact uniform representation value of data cluster A is:

[0021] in, For the compact and uniform representation value of data point cluster A, The number of data points in data point cluster A. Let be the intra-cluster nearest neighbor distance of the j-th data point in data cluster A. Let be the mean intra-cluster distance of the j-th data point in cluster A, and Norm() be the normalization function. The larger the value, the less densely distributed the data points within data cluster A. The larger the value, the more uneven the distribution of data points within data cluster A. Furthermore, within the tree canopy, due to the interlacing of branches and leaves, the LiDAR scan involves multiple reflections, further resulting in a denser and more uniform distribution of point cloud data points within the canopy. In contrast, the distribution of clustered noise points generated by cloud and fog refraction exhibits greater randomness due to refraction angle deviations, leading to a weaker density and uniformity in the distribution. In other words, the density and uniformity of data point distribution within non-forest stand structure noise clusters are relatively weak. The larger and The larger the value, the less dense and uneven the distribution of data points within cluster A. The larger and The larger the value, the more likely data point cluster A is to be a non-forest stand structure noise cluster, or the stronger the noise level of data point cluster A; while The larger and When it is larger, The larger, therefore The larger the value, the more likely the data point cluster A is to be a non-forest stand structure noise cluster, and vice versa. The smaller the value, the less likely data point cluster A is to be a non-stand structure noise cluster, and the more likely it is to be a tree canopy cluster.

[0022] Because the tree canopy contains many branches, its point cloud data distribution exhibits strong edge extension characteristics. Therefore, in this embodiment, region growing is performed on the data points in data point cluster A to obtain all growth edge lines of data point cluster A. The growth edge lines obtained by region growing the data points in data point cluster A can reflect edge extension characteristics. The growth edge lines can also reflect the possibility that data point cluster A belongs to a forest canopy cluster or the possibility that data point cluster A belongs to a non-stand structure noise cluster, which is also a key parameter for determining non-stand structure noise clusters. The specific process of performing region growing on the data points in data point cluster A to obtain all growth edge lines of data point cluster A is as follows: (1) Randomly select a data point from data point cluster A as the initial growth point; (2) Select the point closest to the initial growth point from the candidate growth set as the second growth point; (3) From the candidate growth set of the current growth point, select the point with the smallest angle between its growth direction and the previous growth point and not exceeding the preset angle threshold as the subsequent growth point; (4) Repeat step (3) until the edge of data point cluster A is reached or no data point in the candidate growth set of the current growth point satisfies the condition that the difference in growth direction angle does not exceed the preset angle threshold, and then terminate the current edge growth. In other words, terminate the current edge growth until no other data point in the candidate growth set of the current growth point satisfies the condition that the difference in growth direction angle is less than the preset angle threshold; (5) Mark the points that have been grown or the points that have been visited, and re-select the remaining unvisited points from data point cluster A. A new initial growth point is randomly selected, and steps (2) to (4) are repeated until all points in data point cluster A are visited or grown; (6) all terminated growth paths are summarized, and each terminated growth path is recorded as a growth edge line of data point cluster A; and the candidate growth set of any growth point in data point cluster A is composed of points within the preset neighborhood radius of the growth point that belong to data point cluster A and have not been grown or visited. That is, if the points within the preset neighborhood radius of a certain growth point in data point cluster A are data point 1, data point 2, data point 3 and data point 4, but currently only data point 1 and data point 2 belong to data point cluster A and have not been visited, then the set composed of data point 1 and data point 2 is the candidate growth set of the growth point; unvisited points and ungrown points refer to points that have not been selected by any growth trajectory, have not been marked as processed, and are still in the waiting state. In specific applications, the implementer needs to set a preset angle threshold according to the actual situation. For example, in this embodiment, the preset angle threshold can be set to 90 degrees.

[0023] For example: If data point cluster A contains H data points, and the set of data points in data point cluster A is {P1, P2, ..., P...} H}, P1 is the first data point in data point cluster A, P2 is the second data point in data point cluster A, P HLet P1 be the last data point in data point cluster A. The preset neighborhood radius is R0, and the preset angle threshold is 90 degrees. If P1 is initially selected as the growth point from data point cluster A, and P2 is the closest unvisited point within P1's preset neighborhood radius, then P2 is selected as the second growth point, with the growth path from P1 to P2. If, within P2's preset neighborhood radius, the angle difference between the growth direction from P2 to P3 and the growth direction from P1 to P2 is the smallest and less than 90 degrees, then P3 is selected as the next or third growth point, with the growth path P1→P2→P3. The process continues from P3. If there are no unvisited points within P3's preset neighborhood radius... If the angle difference between the growth direction from P3 to an unvisited point in data point cluster A and the growth direction from P2 to P3 is not less than 90 degrees among all unvisited points in data point cluster A within the preset neighborhood radius of P3, then growth is terminated, and a growth edge line P1→P2→P3 is obtained for data point cluster A. Then, a data point is selected from the unvisited points in data point cluster A as the initial seed point again. If P4 is selected as the initial seed point, then the growth process when P1 is the initial growth point is repeated until all points in data point cluster A are visited or grown, and all growth edge lines of data point cluster A are summarized. The growth direction from P1 to P2 refers to the angle of the vector from P1 to P2, and the growth directions between other data points are similar.

[0024] Since most branches in the canopy are continuous and smooth during growth, with only some branches exhibiting significant directional changes, the linear distribution of the growth edge of the canopy cluster is relatively good. In other words, the longer the overall growth edge of the data point cluster, the better the linear distribution, and the more pronounced the canopy branch characteristics of the corresponding data point cluster. Conversely, the shorter the overall growth edge of the data point cluster, the worse the linearity, and the more the corresponding data point cluster conforms to the clustering characteristics of randomly distributed refractive noise, or the more likely it is a non-stand structure noise cluster. The length of the growth edge can be measured by the sum of the distances between adjacent data points on the growth edge, and the linearity of the growth edge can be measured by the slope between adjacent data points on the growth edge. Therefore, the growth edge... The line length and the slope between adjacent data points on the growth edge line can reflect the probability that data point cluster A is a noise cluster. Therefore, in this embodiment, after obtaining all the growth edge lines of data point cluster A, the length of each growth edge line of data point cluster A and the slope between adjacent data points on the growth edge line of data point cluster A are obtained. Based on the average length of all the growth edge lines of data point cluster A and the slope between adjacent data points on the growth edge line of data point cluster A, the regional growth characterization value of data point cluster A is obtained. The regional growth characterization value can reflect the probability that data point cluster A is a non-stand structure noise cluster, and it is also a key parameter reflecting whether data point cluster A is a non-stand structure noise cluster. The specific process of obtaining the regional growth characterization value of data point cluster A is as follows: Calculate the average length of all growth edge lines in data point cluster A, and perform a negative correlation mapping on the average length of all growth edge lines in data point cluster A. Record the mapping result as the first growth index value. The length of any growth edge line is the sum of the distances between adjacent data points on that growth edge line. For example, if a growth edge line is from data point G0 to data point G1, and then from data point G1 to data point G2, then the length of that growth edge line is the sum of the distance between data points G0 and G1 and the distance between data points G1 and G2. The length of each growth edge line in data point cluster A is calculated as the sum of the distances between adjacent data points. The sequence of slopes between data points is denoted as the slope sequence of the corresponding growth edge line. The g-th slope in the slope sequence of any growth edge line is the slope between the g-th data point and the (g+1)-th data point on that growth edge line. The number of slopes in the slope sequence of any growth edge line is the number of data points on that growth edge line minus 1. The mean of the absolute values ​​of the differences between adjacent slopes in the slope sequence of the growth edge lines of data point cluster A is calculated and denoted as the change characteristic value of the corresponding growth edge line. The mean of the change characteristic values ​​of all growth edge lines of data point cluster A is denoted as the region growth characteristic value of data point cluster A. The expression for the region growth characteristic value of data point cluster A is:

[0025] in, The region growth characterization value for data point cluster A, Let A be the number of growth edge lines of data point cluster A. Let c be the length of the c-th growth edge line of data point cluster A. This is the value representing the degree of change of the c-th growth edge line of data point cluster A; ,in, Let be the number of slopes in the slope sequence of the c-th growth edge line of data point cluster A. Let be the s-th slope in the slope sequence of the c-th growth edge line of data point cluster A. Let be the (s+1)th slope in the slope sequence of the c-th growth edge line of data point cluster A. The larger the value, the greater the degree of change between the corresponding data point positions; Characterizes the overall growth edge length of data point cluster A. The larger the value, the longer the overall growth edge line of data point cluster A, the more obvious the canopy and branch characteristics of data point cluster A, and the less likely data point cluster A is a non-stand structure noise cluster; Characterizes the overall linearity of the growth edge line of data point cluster A. The smaller the value, the better the overall linearity of the growth edge of data cluster A. This indicates that the canopy and branch characteristics of data cluster A are more pronounced, and the likelihood that data cluster A is a noise cluster outside the forest stand structure is lower. The larger the value, the more complex the distribution of the overall edges of data point cluster A, the more it conforms to the clustering characteristics of random distribution of refractive noise, and the higher the degree of clustered data noise in data point cluster A; because The larger and The smaller, The smaller, therefore The smaller the value, the more pronounced the canopy and branch characteristics of data point cluster A, and the less likely data point cluster A is a non-stand structure noise cluster; conversely, the larger the value, the more likely it is. The larger the value, the more obvious the clustered noise characteristics of data point cluster A, and the greater the possibility that data point cluster A is a non-forest stand structure noise cluster.

[0026] Since both the compact uniformity characterization value and the regional growth characterization value can reflect the possibility that data cluster A is a non-stand structure noise cluster, this embodiment, after obtaining the compact uniformity characterization value and the regional growth characterization value of data cluster A, fuses the compact uniformity characterization value and the regional growth characterization value of data cluster A to obtain the cluster noise characterization value of data cluster A. In this embodiment, the sum of the compact uniformity characterization value and the regional growth characterization value of data cluster A is recorded as the cluster noise characterization value of data cluster A. The larger the cluster noise characterization value of data cluster A, the greater the possibility that data cluster A is a non-stand structure noise cluster.

[0027] Therefore, this embodiment can obtain the cluster noise characterization value of each data point cluster through the above process; after obtaining the cluster noise characterization value of the data point cluster, non-forest stand structure noise clusters are screened based on the cluster noise characterization value of the data point cluster. Specifically, for any data point cluster, it is determined whether the cluster noise characterization value of the data point cluster is greater than the preset cluster noise judgment threshold. If it is greater, the data point cluster is determined to be a non-forest stand structure noise cluster, and the points in the non-forest stand structure noise cluster are non-forest stand structure noise points. Otherwise, it indicates that the data point cluster does not belong to the non-forest stand structure noise cluster, but belongs to the forest canopy cluster, and the points in the forest canopy cluster belong to the forest canopy. In practical applications, implementers need to set a preset cluster noise judgment threshold based on the actual situation such as the range of cluster noise characterization values. For example, in this embodiment, the tree canopy confidence feature values ​​of forest canopy clusters and non-forest stand structure noise clusters can usually be divided into the two ends of the cluster noise characterization value range through the above process. Therefore, in order to better identify non-forest stand structure noise clusters, this embodiment can use the median value of the cluster noise characterization value range, 1, as the preset cluster noise judgment threshold.

[0028] Therefore, this embodiment can obtain isolated non-forest stand structure noise points and non-forest stand structure noise clusters through the above process.

[0029] Step S003: Remove isolated non-stand structure noise points and non-stand structure noise clusters from the original digital surface model and the original digital terrain model to obtain the target digital surface model and the target digital terrain model. Obtain the stock volume of the plantation to be measured based on the target digital surface model and the target digital terrain model.

[0030] In this embodiment, after obtaining isolated non-stand structure noise points and non-stand structure noise clusters, points belonging to isolated non-stand structure noise points and non-stand structure noise clusters in the original digital surface model and original digital terrain model are removed. The original digital surface model and original digital terrain model after removing non-stand structure noise points are respectively denoted as the target digital surface model and the target digital terrain model. Points in isolated non-stand structure noise points and non-stand structure noise clusters are all non-stand structure noise points. Finally, based on the denoised target digital surface model and target digital terrain model, the final CHM (canopy height model) is obtained and denoted as the target CHM. The volume of the plantation to be measured is obtained based on the obtained target CHM. The target CHM is the result of subtracting the target digital terrain model from the target digital surface model. Since the process of obtaining the volume of the plantation from the digital surface model and digital terrain model or CHM of the plantation is known, it will not be described in detail.

[0031] Thus, this embodiment completes the calculation of plantation forest volume. This embodiment achieves the calculation of plantation forest volume by identifying and removing non-stand structure noise, and by using the CHM determined from the digital surface model and digital terrain model after removing non-stand structure noise. This minimizes or eliminates the impact of non-stand structure noise on the accuracy of extracting core parameters such as tree height and canopy structure, and also minimizes non-stand structure noise present in the DSM and DTM, improving the accuracy of plantation forest volume calculation. Furthermore, in this embodiment, only specific numerical values ​​are used for parameters in the calculation; dimensions are not involved.

[0032] In summary, this embodiment first obtains the original model data point set corresponding to the plantation to be measured; then, based on the number of points within a preset neighborhood radius of each data point in the original model data point set and the distance between each data point and each point within the preset neighborhood radius, the canopy confidence feature value of each data point is obtained. Based on the canopy confidence feature value, isolated non-stand structure noise points are screened out. The isolated non-stand structure noise points in the original model data point set are removed to obtain the set to be processed. The data points are clustered according to the distance between the data points in the set to be processed to obtain each data point cluster. Based on the nearest neighbor distance within each data point cluster and the results of regional growth of each data point cluster, the cluster noise characterization value of each data point cluster is obtained. Non-stand structure noise clusters are screened according to the cluster noise characterization value. Finally, points belonging to isolated non-stand structure noise points and non-stand structure noise clusters in the original digital surface model and the original digital terrain model are removed to obtain the target digital surface model and the target digital terrain model. The volume of the plantation to be measured is obtained based on the target digital surface model and the target digital terrain model. Furthermore, this embodiment identifies isolated non-stand structural noise points and non-stand structural noise clusters based on canopy confidence feature values ​​and cluster noise point characterization values. It also determines the plantation stock volume to be measured based on the target digital surface model and target digital terrain model after removing isolated non-stand structural noise points and non-stand structural noise clusters. This can reduce or eliminate the interference of environmental disturbances on the accuracy of plantation biomass measurement, thereby improving the accuracy of plantation stock volume measurement.

[0033] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. An intelligent method for calculating the volume of a plantation based on a laser radar point cloud, characterized by, The method includes the following steps: Obtain the set of original model data points corresponding to the artificial forest to be measured. The set of original model data points consists of data points from the original digital surface model and the original digital terrain model. Based on the number of points within a preset neighborhood radius of each data point in the original model data point set and the distance between each data point and each point within the preset neighborhood radius, the canopy confidence feature value of each data point is obtained. Isolated non-stand structure noise points are screened out based on the canopy confidence feature value. Isolated non-stand structure noise points in the original model data point set are removed to obtain a set to be processed. The data points are clustered according to the distance between the data points in the set to be processed to obtain each data point cluster. Based on the nearest neighbor distance within each data point cluster and the results of region growing for each data point cluster, the cluster noise characterization value of each data point cluster is obtained. Non-stand structure noise clusters are screened out based on the cluster noise characterization value. Points belonging to isolated non-stand structure noise points and non-stand structure noise clusters in the original digital surface model and original digital terrain model are removed to obtain the target digital surface model and target digital terrain model. The stock volume of the plantation to be measured is obtained based on the target digital surface model and target digital terrain model.

2. The intelligent method for calculating the stock volume of planted forests based on lidar point clouds as described in claim 1, characterized in that, Methods for obtaining the preset neighborhood radius include: From the set nearest neighbor distances of all data points belonging to the original model data point set, the largest set nearest neighbor distance is selected as the preset neighborhood radius. The set nearest neighbor distance of a data point refers to the minimum value of the Euclidean distance between the corresponding data point and other data points in the original model data point set.

3. The intelligent method for calculating the stock volume of planted forests based on lidar point clouds as described in claim 1, characterized in that, Methods for obtaining tree canopy confidence feature values ​​include: The normalized result of the number of data points within the preset neighborhood radius of each data point in the original model data point set is recorded as the neighborhood number measure of the corresponding data point; the inverse normalized result of the mean Euclidean distance between each data point in the original model data point set and each data point within its preset neighborhood radius is recorded as the distance representation value of the corresponding data point. The sum of the neighborhood count of each data point and the distance representation of the corresponding data point is denoted as the canopy confidence feature value of the corresponding data point.

4. The intelligent method for calculating the stock volume of planted forests based on lidar point clouds as described in claim 1, characterized in that, Data points in the original model data point set whose tree canopy confidence feature value is less than the preset isolated noise threshold are isolated non-forest stand structure noise points.

5. The intelligent method for calculating the stock volume of planted forests based on lidar point clouds as described in claim 1, characterized in that, Methods for obtaining cluster noise representation values ​​for each data point cluster include: For any data point cluster: Based on the mean of the intra-cluster nearest neighbor distances of all data points in the data point cluster and the mean of the absolute values ​​of the differences between the intra-cluster nearest neighbor distances of data points in the data point cluster and the intra-cluster nearest neighbor distances of other data points in the cluster, a compact uniform characterization value for the data point cluster is obtained. Region growth is performed on the data points in the data point cluster to obtain the growth edge lines of the data point cluster. Based on the mean length of all growth edge lines of the data point cluster and the slope between adjacent data points on the growth edge lines, a region growth characterization value for the data point cluster is obtained. The sum of the compact uniform characterization value and the region growth characterization value is recorded as the cluster noise characterization value of the data point cluster.

6. The intelligent method for calculating the stock volume of planted forests based on lidar point clouds as described in claim 5, characterized in that, The method for obtaining the compact and uniform representation values ​​of the data point clusters includes: The mean of the intra-cluster nearest neighbor distances of all data points in the data point cluster is denoted as the first intra-cluster distribution characteristic value; the mean of the absolute values ​​of the differences between the intra-cluster nearest neighbor distances of each data point in the data point cluster and the intra-cluster nearest neighbor distances of all other data points in the cluster is denoted as the mean intra-cluster distance of the corresponding data point; the mean of the mean intra-cluster distances of all data points in the data point cluster is denoted as the second intra-cluster distribution characteristic value; and the normalized result of adding the first intra-cluster distribution characteristic value and the second intra-cluster distribution characteristic value is denoted as the compact uniformity characterization value of the data point cluster.

7. The intelligent method for calculating the stock volume of planted forests based on lidar point clouds as described in claim 5, characterized in that, The method for obtaining the growth edge line of the data point cluster includes: A data point is randomly selected from the data point cluster as the initial growth point. Within the candidate growth set of the initial growth point, the point closest to it is selected as the second growth point. Iteratively, from the candidate growth set of the current growth point, the point with the smallest difference in growth direction angle from the previous growth point, and not exceeding a preset angle threshold, is selected as the next growth point. This process continues until there are no other data points in the candidate growth set of the current growth point, or no data point in the candidate growth set of the current growth point satisfies the condition that the difference in growth direction angle is less than the preset angle threshold. Growth then stops, resulting in a growth edge line of the data point cluster. From the unvisited data points in the data point cluster, a new initial growth point is randomly selected, and the above growth process is repeated until all data points in the data point cluster have been visited. All growth edge lines of the data point cluster are then statistically analyzed. The candidate growth set of any growth point consists of all unvisited points within a preset neighborhood radius of the growth point that belong to the data point cluster.

8. The intelligent method for calculating the stock volume of planted forests based on lidar point clouds as described in claim 5, characterized in that, The method for obtaining the region growth characterization value of the data point cluster includes: The negative correlation mapping result of the mean length of all growth edge lines of the data point cluster is recorded as the first growth index value. The sequence of slopes between adjacent data points on each growth edge line is recorded as the slope sequence of the corresponding growth edge line. The mean of the absolute values ​​of the differences between adjacent slopes in the slope sequence of the growth edge line is recorded as the change degree characterization value of the corresponding growth edge line. The mean of the change degree characterization values ​​of all growth edge lines of the data point cluster is recorded as the regional growth characterization value of the data point cluster.

9. The intelligent method for calculating the stock volume of planted forests based on lidar point clouds as described in claim 1, characterized in that, The cluster noise characterization value of the non-forest stand structure noise cluster is greater than the preset cluster noise judgment threshold.

Images