A bottom-up point cloud single tree segmentation method and system

Abstract

The present invention relates to the technical field of three-dimensional point cloud, and discloses a bottom-up point cloud single tree segmentation method and system, wherein the rubber forest in the full leaf stage is collected by an unmanned aerial vehicle laser radar to form the initial data of the full leaf stage, and the rubber forest in the deciduous stage is collected to form the initial data of the deciduous stage; the initial data of the full leaf stage and the initial data of the deciduous stage are pre-processed respectively to obtain the point cloud data of the full leaf stage and the point cloud data of the deciduous stage; the two-phase point cloud data are divided into blocks according to the position to form multiple groups of rubber forest point cloud data; vertical stratification is performed from bottom to top for each group of deciduous rubber forest point cloud data, and low vegetation under the forest is removed; a bottom-up single tree segmentation method is adopted to realize the single tree segmentation of the deciduous rubber tree; the point cloud data of the full leaf stage is based on the single tree segmentation of the deciduous stage, and the basic skeleton point cloud of the single tree in the deciduous stage is used as the initial calculation data to realize the single tree segmentation of the rubber tree in the full leaf stage. The single tree point cloud data formed by the present invention is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional point clouds, and in particular to a bottom-up point cloud single tree segmentation method and system. Background Art

[0002] Natural rubber is an indispensable strategic material and scarce resource for my country's national defense and economic construction. It plays an irreplaceable role in the manufacturing of high-end equipment such as military, aerospace, and ocean. Among them, the Brazilian rubber tree (Hevea brasiliensis) is the world's main source of natural rubber. It likes high temperature and high humidity and is not cold-resistant. It is mainly planted in Hainan Province, southern Yunnan, western Guangdong, Guangxi and Fujian. Because my country's rubber planting area is located on the northern edge of the equator with low temperatures and often suffers from natural disasters such as cold and drought, and typhoon disasters frequently occur in coastal areas such as Hainan, making it difficult for my country's natural rubber production to reach an ideal state.

[0003] In order to effectively deal with natural disasters and minimize the loss of natural rubber production, people have made various attempts. For example, by installing cameras and connecting them to remote monitoring systems, the growth status of rubber forests can be monitored at all times; by using drones to scan the outline of rubber forests to determine the overall growth of rubber forests.

[0004] However, various existing methods including the above-mentioned trial methods, although they can complete the overall monitoring of the rubber forest, cannot effectively resist the yield loss caused by various natural disasters to the rubber forest. The reason is that the rubber forest is only monitored from the overall outline, and the growth of each rubber tree cannot be monitored in time. When the rubber tree grows poorly and shows obvious changes in the overall outline of the rubber forest, it is often impossible to recover the losses that have already occurred. At the same time, it is also difficult to prevent the decline of the rubber forest, which is not conducive to taking rescue measures in time.

[0005] Although some people have thought of monitoring the growth of each rubber tree, it is almost impossible to accurately monitor it from a distance because the shape of the rubber tree changes greatly during its growth cycle. When it is at its most lush, it can even make the entire rubber forest become one piece. If you enter the rubber forest to monitor each rubber tree, you need to go there regularly to check, which is difficult to achieve ideal conditions in terms of labor costs and timeliness of feedback. Summary of the invention

[0006] The present invention aims to provide a bottom-up point cloud single tree segmentation method to solve the current problem that the actual growth conditions of rubber forests cannot be remotely monitored.

[0007] To solve the above problems, the present invention adopts the following technical solutions:

[0008] A bottom-up point cloud single tree segmentation method includes the following steps:

[0009] Step 1: using the UAV laser radar, collect the initial data of the rubber forest in the full leaf stage to form the full leaf stage, and collect the initial data of the rubber forest in the deciduous stage to form the deciduous stage;

[0010] Step 2, preprocessing the initial data of the full leaf stage and the initial data of the leaf falling stage respectively to obtain the point cloud data of the full leaf stage and the point cloud data of the leaf falling stage;

[0011] Step 3, dividing the point cloud data of the full leaf stage and the point cloud data of the leaf falling stage into blocks according to the positions to form multiple groups of block data of the full leaf stage and block data of the leaf falling stage;

[0012] Step 4: vertically stratify each set of deciduous period block data from bottom to top, and divide each set of deciduous period block data into understory vegetation layer, trunk layer and canopy layer according to the variation law of point cloud density in the vertical structure direction of the forest, and remove the understory low vegetation data contained in the understory vegetation layer according to the preset understory screening strategy to obtain deciduous period block processing data;

[0013] Step 5, extracting the basic skeleton point cloud data of the rubber tree in the leaf-falling stage by using a hierarchical clustering method from the leaf-falling stage block processing data as the leaf-falling stage skeleton data;

[0014] Step 6, for the deciduous period skeleton data, a bottom-up segmentation method is used to implement the single tree segmentation of the deciduous period rubber tree, to obtain the deciduous period single tree trunk data and the deciduous period single tree crown data, and the deciduous period single tree trunk data and the deciduous period single tree crown data are combined to form the deciduous period single tree point cloud data;

[0015] Step 7, using the deciduous period skeleton data as the full leaf period skeleton data of the full leaf period block data corresponding to the group of deciduous period block data, and dividing each group of full leaf period block data from bottom to top into full leaf period single tree trunk data and full leaf period single tree crown data in the same manner as step 6; and combining the full leaf period single tree trunk data and the full leaf period single tree crown data to form full leaf period single tree point cloud data;

[0016] Step 8: Combine the single tree trunk data in the deciduous period with the single tree trunk data in the full leaf period to form the single tree trunk data, combine the single tree crown data in the deciduous period with the single tree crown data in the full leaf period to form the single tree crown data; and combine the single tree trunk data and the single tree crown data to form the single tree point cloud data.

[0017] The principles and advantages of this solution are:

[0018] Compared with the prior art, only the rubber forest can be monitored as a whole, but the growth of individual trees cannot be accurately divided; the present invention utilizes the growth characteristics of rubber trees, combines the data of the peak leaf period and the deciduous period, accurately divides and restores the individual tree data, and makes up for the difficulty of not being able to monitor the rubber forest from the individual tree data. Through the present invention, the growth of each rubber tree can be monitored in real time, and the rubber trees can be picked in time, which can effectively avoid the product losses caused by natural disasters.

[0019] More importantly, the present invention can reasonably divide the vertical layered content of a single tree through a bottom-up single tree segmentation method. When forming single tree point cloud data, the amount of calculation can be minimized while ensuring accuracy. Under the same computing power resources, the obtained single tree point cloud data is more consistent with the actual growth of the single tree. Through the present invention, the prerequisite is provided for effectively monitoring the rubber forest and truly restoring the growth of the rubber forest.

[0020] Preferably, as an improvement, the peak leaf period refers to the period when the crown of the rubber tree grows vigorously and more than 80% of the new leaves grow into old leaves; the deciduous period refers to the period when the crown of the rubber tree withers and more than 80% of the leaves fall.

[0021] Effect: The peak leaf period and the leaf fall period are obvious growth characteristics of rubber trees found through research. The accurate identification of the peak leaf period and the leaf fall period is helpful to improve the accuracy and true restoration of the formed single tree point cloud data. At the same time, the study found that when the temperature is below 15°, the oak tree begins to shed its leaves. Therefore, the leaf fall period also refers to the period of at least half a month when the temperature is below 15° and remains below 15°. If the time period after 80% of the leaves fall coincides with the time period when the temperature is below 15°, the overlapping time period is the actual leaf fall period. If they do not overlap, the time period from 80% of the leaves falling to the time when the leaves fall completely is the actual leaf fall period. With this setting, the leaf fall period can be identified more timely and accurately through the point cloud data collected by the drone under the premise of limited computing power, and the leaf fall period data collection can be completed.

[0022] Preferably, as an improvement, in step 5, the basic skeleton point cloud data of the rubber tree in the deciduous period is extracted using a hierarchical clustering method from the deciduous period block processing data as the deciduous period skeleton data, comprising the following steps:

[0023] Step 51, stratifying the point cloud, stratifying the leaf-falling period block-processed data from bottom to top according to a certain height;

[0024] Step 52, using a clustering algorithm for each layer of point cloud, clustering the point cloud of the branches and trunks of the rubber tree to form multiple groups of point cloud clusters;

[0025] Step 53, obtain the center point of the point cloud in each point cloud cluster as the basic skeleton point of the cluster, extract all the basic skeleton points in a set of deciduous period block processing data to form the deciduous period skeleton data of the block area, and extract all the basic skeleton points in all the block areas to form the deciduous period skeleton data of the forest area.

[0026] Effect: In this scheme, a block area contains at least one rubber tree. By processing the block data of the deciduous period, the hierarchical clustering algorithm can quickly form point cloud clusters, and then the basic skeleton points can be quickly obtained. It is convenient for single tree segmentation in the subsequent step 6. If there are several rubber trees in a block area, several groups of single tree data can be obtained.

[0027] By adopting the above method, the unit area can be quickly segmented according to the layering method to form multiple sets of data from bottom to top, and then the corresponding basic skeleton points can be quickly formed in each height layer, and the skeleton data formed by the basic skeleton points of each layer in the unit area can be extracted. By calculating all the block areas, the skeleton data of the entire rubber forest area can be formed.

[0028] Preferably, as an improvement, in step 6, a bottom-up segmentation method is used to achieve single-tree segmentation of rubber trees in the deciduous stage, comprising the following steps:

[0029] Step 61, using the skeleton data of the leaf-falling period as the basic data, taking the lowest point of the point cloud height as the initial seed point, and starting to grow upward;

[0030] Step 62, setting two horizontal distance thresholds according to the planting rules of the rubber forest, starting from the initial seed point, calculating the distance and horizontal distance between the point clouds, and classifying the points with the smallest distance between the two points and less than the horizontal distance threshold as the seed points of the same tree;

[0031] Step 63, classifying all basic skeleton points to complete the extraction of the basic skeleton point cloud of a single tree in the entire rubber forest, and forming multiple groups of basic skeleton seed point sets;

[0032] Step 64, classifying all the deciduous point cloud data according to the one-to-one correspondence of the clusters in step 52, so as to achieve the single tree segmentation of the deciduous rubber trees.

[0033] Effect: In the peak leaf stage, because of its luxuriant branches and leaves and large crown width, it is very easy for two trees to form adhesions. The skeleton data of the deciduous stage provides a basis for the division of the single tree point cloud data in the peak leaf stage. Using the improved regional growing algorithm, the skeleton data of the deciduous stage is used as the basic data, and the lowest point of the point cloud height is used as the initial seed point to start growing upward; two horizontal distance thresholds are set according to the preset planting rules of the rubber forest, starting from the initial seed point, the Euclidean distance and horizontal distance between the point clouds are calculated, and the points with the smallest distance between the two points and less than the horizontal distance threshold are classified as seed points of the same tree; in each layer of skeleton data, it can quickly segment and form point cloud sets belonging to different trees, and according to the coordinate position, the point cloud sets formed by seed points of different heights on the same coordinate are gathered together to form the point cloud set corresponding to this rubber tree, and then all the deciduous point cloud data are classified according to the multiple groups of clusters corresponding to each rubber tree, so as to achieve the single tree segmentation of the deciduous rubber tree.

[0034] Because a tree has a continuous and complete growth cycle in an adjacent leaf-full period and leaf-falling period, the trunk position of the same rubber tree will not change, and the chest radius range of the trunk will not change too much. Therefore, using the trunk data of the leaf-falling period for calculation not only does not affect the accuracy, but also saves calculation steps.

[0035] Preferably, as an improvement, in step 5, a hierarchical DBSCAN clustering algorithm is used to extract basic skeleton points of rubber trees in the deciduous stage.

[0036] Effect: The hierarchical DBSCAN clustering algorithm itself is not conducive to finding the center point, but in this solution, hierarchical clustering can be quickly implemented, and the basic skeleton points can be obtained from the point cloud cluster cluster by defining the basic skeleton points. It can save calculation while meeting the accuracy requirements.

[0037] Preferably, as an improvement, when using the hierarchical DBSCAN clustering algorithm to extract the basic skeleton points of rubber trees in the deciduous period, the axis of the trunk is defined to be consistent with the crown, and the points on the axis are first determined. Then, based on the symmetry setting, the positions of the lateral end points of the trunk and the crown are determined. Then, the positions of the edge points of the trunk are determined based on the breast radius range, and the positions of the edge points of the crown are determined based on the crown width to form basic skeleton points.

[0038] The setting of basic skeleton points provides the premise for the accurate division of the trunk layer and the canopy layer, which can effectively make up for the shortcomings of the DBSCAN algorithm.

[0039] Preferably, as an improvement, when the peak leaf period block data is segmented using a bottom-up improved region growing algorithm based on basic skeleton points, the breast radius range is dynamically adjusted, and when the position of the trunk edge point is asymmetric, it is corrected according to the actual point cloud position collected.

[0040] Effect: According to the actual growth of rubber trees, the chest radius range is dynamically adjusted to simulate and restore the trunk thickness of the rubber trees. The edge point position corresponding to the single tree point cloud data is corrected by determining the edge position of the actual point cloud, so that the formed single tree point cloud data is more consistent with the actual shape and growth of the rubber tree, increasing the authenticity and restoration, and providing a basic premise for more accurate monitoring of the growth of rubber trees.

[0041] Preferably, as an improvement, in step 8, the single tree trunk data in the deciduous period and the single tree trunk data in the full leaf period are combined to form the single tree trunk data, including the following contents:

[0042] The single tree trunk data in the deciduous period and the single tree trunk data in the full leaf period are overlapped, and the breast radius range of the deciduous period trunk and the full leaf period trunk are compared by comparing the position coordinates of the trunk edge points. If the breast radius range of the deciduous period trunk is greater than or equal to the breast radius range of the full leaf period trunk, the single tree trunk data in the deciduous period is regarded as the single tree trunk data;

[0043] If the breast radius range of the trunk in the deciduous period is smaller than the breast radius range of the trunk in the full leaf period, the tree will be marked as diseased.

[0044] Effect: The changes in the chest radius of the trunk within a growth cycle are used to obtain the changes in the trunk thickness. Rubber trees corresponding to the abnormal data are marked as diseased trees, which can timely reflect the actual growth status of each rubber tree.

[0045] Preferably, as an improvement, in step 8, the single tree crown data in the deciduous period and the single tree crown data in the full leaf period are combined to form the single tree crown data, including the following contents:

[0046] The crown data of a single tree in the deciduous period and the crown data of a single tree in the full leaf period are overlapped, and the crown width length ranges of the crowns in the deciduous period and the full leaf period are compared by comparing the position coordinates of the crown edge points. If the crown width length range of the crown in the full leaf period is larger than that of the crown in the deciduous period, the data set formed by shrinking the crown width length of the edge position points of the crown data of the single tree in the full leaf period by more than one third is used as the crown data of the single tree;

[0047] If the crown width length range of the crown layer in the peak leaf period is less than or equal to the crown width length range of the trunk layer in the deciduous period, the tree will be marked as diseased.

[0048] Effect: The changes in the crown length within a growth cycle are used to obtain the changes in the crown branches and leaves. The rubber trees corresponding to the abnormal data are marked as diseased trees, which can timely reflect the actual growth status of each rubber tree.

[0049] The present invention also provides a segmentation system based on the above segmentation method, using the above bottom-up point cloud single tree segmentation method. Through this system, the three-dimensional point cloud data of the rubber forest that has been collected can be quickly segmented into single trees by the above segmentation method in any intelligent terminal, and the growth of each rubber tree can be restored more realistically, providing a prerequisite for accurate monitoring of the entire rubber forest, and the rubber trees can be picked in time according to their actual growth conditions, which can effectively avoid the yield losses caused by natural disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The figure is a schematic diagram for comparing a rubber forest in the full leaf stage and a rubber forest in the deciduous stage in Example 1 of the present invention.

[0051] Figure 2 This is a flow chart of Example 1 of the present invention.

[0052] Figure 3 This is a schematic diagram of the tree growth process in the improved region growing algorithm in Example 5 of the present invention. DETAILED DESCRIPTION

[0053] The following is further described in detail through specific implementation methods:

[0054] The bottom-up point cloud single tree segmentation method in the present invention combines the fallen leaf and full leaf data to improve the integrity of the rubber tree plant type parameters and emphasize the authenticity of the tree point cloud data.

[0055] The present invention is mainly directed to tropical rubber trees, and is used to monitor the growth of rubber forests. In the past, the parameter measurements of rubber trees were all manual operations, such as the height (height under the branches) of trunks, tree height, breast diameter, crown width, crown volume, etc., which were cumbersome to operate, and because different varieties of rubber tree planting methods are different, manual measurement monitoring is inconvenient. This scheme uses unmanned aerial vehicle radar to scan and collect rubber tree forests in the same growth cycle during the deciduous period and the full leaf period to obtain two forest point cloud data. By point cloud cluster analysis, in the forest point cloud data, first find the center point of each tree, identify the basic skeleton point cloud data of rubber tree, and then segment and identify the individual rubber tree point cloud data from bottom to top for the basic skeleton point cloud data.

[0056] Among them, the single tree point cloud data for segmentation and identification is formed by combining the single tree point cloud data in the dead leaf period and the single tree point cloud data in the full leaf period, which makes up for the shortcomings that the trunk data cannot be collected due to leaves blocking the full leaf period and the crown data cannot be collected due to leaves falling in the dead leaf period. The formed single tree point cloud data can more realistically simulate the actual growth conditions of the corresponding trees, making it easier to monitor tree growth more accurately.

[0057] like Figure 1As shown, the peak leaf period and the leaf falling period of the rubber forest can be clearly distinguished from the appearance. The present invention can accurately determine the peak leaf period and the leaf falling period through the point cloud data collected by the drone.

[0058] For the deciduous tree trunks, the data of individual trees is segmented from bottom to top. The trunk is first found and then the crown is found from bottom to top to segment the data of individual trees. When segmenting, the axis of the trunk is defined to be consistent with the crown. During the segmentation process, the point on the axis is first determined, and then the position of the horizontal end points of the trunk and crown is determined according to the symmetry setting. Then the position of the entire trunk is determined according to the chest radius range. Similarly, the crown position is determined according to the crown width and its changes. When there is asymmetry, the actual point cloud position collected will be judged, and the edge position of the trunk and crown on the corresponding side will be corrected (correction for actual asymmetric trees).

[0059] Combining the data from the two periods, a more complete point cloud data set of the tree is formed. The height under the branches remains unchanged; the tree height, crown width and breast diameter are changing. The breast diameter has different values ​​measured according to different varieties, so it can be directly increased.

[0060] Based on drone monitoring, this solution can analyze point cloud data to truly simulate and restore the growth of a single tree, guide mixed planting, rationally use land, realize the planting of multiple different varieties of rubber forests, and improve economic benefits.

[0061] Natural rubber is an indispensable strategic material and scarce resource for my country's national defense and economic construction. It plays an irreplaceable role in the manufacturing of high-end equipment such as military, aerospace, and ocean. The Brazilian rubber tree (Hevea brasiliensis) is the main source of natural rubber in the world. It likes high temperature, high humidity and is not cold-resistant. It is mainly planted in Hainan Province, southern Yunnan, western Guangdong, Guangxi and Fujian in my country. Since my country's rubber planting area is located on the northern edge of the equator with low temperature and often suffers from natural disasters such as cold and drought, and typhoon disasters frequently occur in coastal areas such as Hainan, it is difficult for my country's natural rubber production to reach an ideal state. In order to effectively respond to natural disasters and minimize the loss of natural rubber production, people have made various attempts. For example, by installing a camera and a remote monitoring system connected to it, the growth status of the rubber forest is monitored at all times; by scanning the outline of the rubber forest by drone, the overall growth of the rubber forest can be judged. However, various existing methods, including the above trial methods, although they can complete the overall monitoring of the rubber forest, cannot effectively resist the loss of production caused by various natural disasters to the rubber forest. The reason why the existing technology can only monitor the whole tree and cannot monitor each rubber tree one by one is that the rubber tree also has obvious leaf-blooming and leaf-falling periods, such as Figure 1As shown in the figure, the appearance characteristics of rubber trees in the two time periods of full leaf period and leaf fall period are quite different due to the canopy obstruction. It is now almost difficult to accurately distinguish each rubber tree without human intervention and only rely on computer settings. In addition, the specific time periods and appearance characteristics of rubber trees in the full leaf period and leaf fall period will also change in different planting environments, which invisibly increases the difficulty of single tree division.

[0062] Embodiment 1

[0063] like Figure 2 As shown, in this embodiment, three-dimensional point cloud collection and single tree point cloud segmentation of the rubber forest include the following steps:

[0064] S1, using UAV laser radar, collects the initial data of the rubber forest in the full leaf period to form the full leaf period, and collects the initial data of the rubber forest in the deciduous period to form the deciduous period;

[0065] S2, preprocessing the initial data of the full leaf stage and the initial data of the leaf falling stage respectively to obtain the point cloud data of the full leaf stage and the point cloud data of the leaf falling stage;

[0066] S3, dividing the two phases of point cloud data into blocks according to the locations to form multiple groups of rubber forest point cloud data;

[0067] S4, vertically stratify each set of point cloud data of rubber forests in the deciduous period from bottom to top, and divide each set of point cloud data of the deciduous period into understory vegetation layer, trunk layer and canopy layer according to the variation law of point cloud density in the vertical structure direction of the forest, and remove the low vegetation under the forest;

[0068] S5, using a hierarchical clustering method to extract the basic skeleton point cloud data of the deciduous rubber tree forest after removing the understory vegetation layer as the calculation data for initial single tree segmentation;

[0069] First, the point cloud is layered, and the rubber forest is layered from bottom to top according to a certain height;

[0070] Then, a clustering algorithm is used for each layer of point cloud to cluster the point cloud of the rubber tree branches and trunks to form multiple groups of point cloud clusters;

[0071] Finally, the center point of the point cloud in each cluster is taken as the basic skeleton point of the cluster, and all the skeleton points are extracted to form the basic skeleton point cloud data of rubber trees in the entire forest area.

[0072] S6, finally, the single tree segmentation method from bottom to top is used to achieve the single tree segmentation of rubber trees in the deciduous period;

[0073] First, the basic skeleton point cloud of the rubber forest is used as the basic data, and the lowest point of the point cloud is used as the initial seed point to start growing upwards;

[0074] Then, two horizontal distance thresholds are set according to the planting rules of rubber forests. Starting from the initial seed point, the distance and horizontal distance between point clouds are calculated, and the points with the smallest distance between the two points and less than the horizontal distance threshold are classified as seed points of the same tree.

[0075] Afterwards, all the basic skeleton point clouds are classified according to the above method to complete the extraction of the basic single tree skeleton point cloud of the entire rubber forest;

[0076] Finally, according to the one-to-one correspondence of the previous clustering clusters, all the deciduous point cloud data are classified to achieve the single tree segmentation of the deciduous rubber trees.

[0077] S7, in the same manner as S4, vertically split each group of point cloud data of rubber forests in the peak leaf stage into understory vegetation layer data in the peak leaf stage, trunk layer data in the peak leaf stage, and canopy layer data in the peak leaf stage;

[0078] S8, based on the segmentation of single trees in the deciduous period, the point cloud data of the full leaf period uses the basic skeleton point cloud of the single tree in the deciduous period as the initial calculation data to achieve the segmentation of single trees in the full leaf period of rubber trees;

[0079] S9, the point cloud data of individual rubber trees obtained after single tree segmentation in the full leaf period and the leaf fall period are combined and used in the subsequent application of rubber tree plant type parameter extraction. The rubber trees in different growth periods can better and more comprehensively reflect the plant type characteristics of rubber trees.

[0080] Specifically,

[0081] In this embodiment, three-dimensional point cloud collection and single tree point cloud segmentation are performed on the rubber forest according to the following method:

[0082] The first step is to collect the initial data of the full leaf period of the rubber forest in the full leaf period through the existing unmanned aerial vehicle laser radar (LiDAR) technology, and collect the initial data of the full leaf period of the rubber forest in the deciduous period; the full leaf period refers to the period when the crown of the rubber tree grows luxuriantly, and more than 80% of the new leaves grow into old leaves; the deciduous period refers to the period when the crown of the rubber tree withers and more than 80% of the leaves fall. The applicant has found that when the temperature is lower than 15 °, the oak tree begins to shed its leaves. Therefore, the deciduous period also refers to the period of at least half a month when the temperature is lower than 15 ° and remains below 15 °. If the time period after the leaves fall 80% overlaps with the time period when the temperature is lower than 15 °, the overlapping time period is the actual deciduous period. If it does not overlap, the time period from the leaves falling 80% to the leaves falling completely is the actual deciduous period. With this arrangement, the deciduous period can be more timely and accurately identified by the point cloud data collected by the unmanned aerial vehicle under the premise of limited calculation amount, and the deciduous period data collection is completed.

[0083] In the second step, the initial data of the full leaf stage and the initial data of the leaf falling stage are preprocessed to obtain the point cloud data of the full leaf stage and the point cloud data of the leaf falling stage; the preprocessing process is as follows:

[0084] First of all, it is necessary to correct the position error of the point cloud data through flight adjustment to improve the quality of the point cloud data.

[0085] Then, according to the existing technology, the point cloud data is subjected to point cloud denoising, ground filtering and point cloud normalization operations. This part of the work occupies most of the time of data preprocessing, and the quality of the processing results directly affects the quality of subsequent work.

[0086] When the quality of point cloud data is improved, the UAV LiDAR system is affected by the altitude and field of view during data collection, and multiple flight strips are required to complete the data collection work in the operation area. In the process of route design, it is necessary to consider the lateral overlap in order to stitch the point clouds between different flight strips and obtain the complete point cloud data of the operation area. Due to the existence of errors in the UAV LiDAR system, there are different degrees of offset problems between the corresponding points of the same name in adjacent flight strips. Before data collection, this study performed internal and external calibration of the UAV LiDAR system, so that the trajectory position and attitude error (referred to as trajectory attitude error) is the main source of point cloud position error. In order to effectively eliminate the residual error of the UAV LiDAR system caused by the trajectory attitude error, this paper uses the Alundar Platform software to perform rigorous flight strip adjustment processing on the two phases of point cloud data, correct the point cloud position, and improve the point cloud quality.

[0087] After rigorous flight zone adjustment of the point cloud data in the full leaf period and the fall leaf period, the point cloud positions were basically corrected. When the point clouds of the two periods are displayed simultaneously, it can be found that although there are a small amount of point cloud position deviations in the point clouds of different periods, it does not affect the subsequent combined application of the two period data, so no joint adjustment work is performed on the two period point cloud data.

[0088] In this embodiment, vertical stratification is performed from bottom to top for each set of single tree point cloud data in the deciduous period, and the deciduous period understory vegetation layer data, deciduous period trunk layer data, and deciduous period canopy layer data corresponding to each set of single tree point cloud data in the deciduous period are obtained;

[0089] For each set of single tree point cloud data in the deciduous period, vertical stratification is performed from bottom to top, including the following:

[0090] Firstly, for each set of single tree point cloud data in the deciduous period, vertical stratification is performed from bottom to top according to the vertical distribution curve of point cloud density to obtain the understory vegetation layer data and the overstory data in the deciduous period.

[0091] Then, the hierarchical DBSCAN clustering algorithm was used to extract the basic skeleton points of rubber trees in the deciduous period, and the improved bottom-up region growing algorithm was used to segment the deciduous forest data based on the basic skeleton points to obtain the corresponding deciduous trunk layer data and deciduous canopy layer data.

[0092] DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a representative density-based clustering algorithm.

[0093] In this embodiment, the region growing algorithm is improved and the existing algorithm is adopted, which will not be described in detail.

[0094] The present embodiment first distinguishes the deciduous period forest undergrowth layer data and the deciduous period forest upper data, and can quickly screen out the forest undergrowth layer data that do not need to participate in subsequent calculations. For the deciduous period forest upper data, the geometric shape of rubber tree is fully considered, first by determining the basic skeleton point, and then confirming the trunk layer and the canopy layer of the corresponding shape from the basic skeleton point. The whole calculation method has thick and thin, and can improve the calculation accuracy of the trunk layer and the canopy layer as much as possible under the premise of limited calculation amount, and at the same time, effectively save calculation steps.

[0095] like Figure 2 As shown, each group of single tree point cloud data in the peak leaf stage is vertically segmented to form understory vegetation layer data in the peak leaf stage, trunk layer data in the peak leaf stage, and canopy layer data in the peak leaf stage, including the following contents:

[0096] Firstly, for each group of single tree point cloud data in full leaf stage, vertical stratification is performed from bottom to top according to the vertical distribution curve of point cloud density to obtain the understory vegetation layer data and overstory data in full leaf stage.

[0097] Then, the hierarchical DBSCAN clustering algorithm was used to extract the basic skeleton points of rubber trees in full leaf stage, and the improved bottom-up region growing algorithm was used to segment the forest data in full leaf stage based on the basic skeleton points to obtain the corresponding trunk layer data and canopy layer data in full leaf stage.

[0098] The trunk layer and canopy layer data during the peak leaf period can be quickly calculated based on the corresponding data during the leaf fall period.

[0099] In this embodiment, when using the hierarchical DBSCAN clustering algorithm to extract the basic skeleton points of the rubber tree in the deciduous period, the axis of the trunk is defined to be consistent with the crown. The points on the axis are first determined, and then the positions of the lateral end points of the trunk and the crown are determined based on the symmetry setting. Then, the positions of the edge points of the trunk are determined based on the breast radius range, and the positions of the edge points of the crown are determined based on the crown width of the crown to form basic skeleton points.

[0100] The setting of basic skeleton points provides the premise for the accurate division of the trunk layer and the canopy layer.

[0101] When the improved bottom-up region growing algorithm is used to segment the forest data in the peak leaf period based on the basic skeleton points, the breast radius range is dynamically adjusted, and the asymmetric position of the trunk edge points is corrected according to the actual point cloud position collected.

[0102] According to the actual growth of the rubber tree, the breast radius range is dynamically adjusted to simulate and restore the thickness of the rubber tree trunk. The edge point position corresponding to the single tree point cloud data is corrected by determining the edge position of the actual point cloud, so that the formed single tree point cloud data is more consistent with the actual shape and growth of the rubber tree, increasing the authenticity and restoration, and providing a basic premise for more accurate monitoring of the growth of rubber trees.

[0103] The trunk layer data of the deciduous period and the trunk layer data of the full leaf period are combined to form the trunk layer data, which includes the following contents:

[0104] Overlap the trunk layer data of the deciduous period with the trunk layer data of the full-leaf period, and compare the breast radius ranges of the trunk layer of the deciduous period and the trunk layer of the full-leaf period by comparing the position coordinates of the trunk edge points. If the breast radius range of the trunk layer of the deciduous period is greater than or equal to the breast radius range of the trunk layer of the full-leaf period, the trunk layer data of the deciduous period is used as the trunk layer data;

[0105] If the breast radius range of the trunk layer in the deciduous period is smaller than the breast radius range of the trunk layer in the full leaf period, the tree will be marked as diseased.

[0106] The changes in the chest radius of the trunk within a growth cycle and the changes in the trunk thickness are obtained. The rubber trees corresponding to the abnormal data are marked as diseased trees, which can timely reflect the actual growth status of each rubber tree.

[0107] The canopy data of the leaf-falling period and the leaf-full period are combined to form the canopy data, including the following:

[0108] Overlap the canopy data of the deciduous period with the canopy data of the full-leaf period, and compare the canopy length ranges of the deciduous period canopy and the full-leaf period canopy by comparing the position coordinates of the canopy edge points. If the canopy length range of the full-leaf period canopy is larger than that of the deciduous period canopy, then the data set formed by shrinking the edge position points of the full-leaf period canopy data inward by more than one-third of the canopy length is used as the canopy data;

[0109] If the crown width length range of the crown layer in the peak leaf period is less than or equal to the crown width length range of the trunk layer in the deciduous period, the tree will be marked as diseased.

[0110] The changes in the length of the crown within a growth cycle and the changes in the branches and leaves of the crown are obtained. The rubber trees corresponding to the abnormal data are marked as diseased trees, which can timely reflect the actual growth status of each rubber tree.

[0111] Compared with the prior art, only the rubber forest can be monitored as a whole, but the growth of individual trees cannot be accurately divided; the present invention utilizes the growth characteristics of rubber trees, combines the data of the peak leaf period and the deciduous period, accurately divides and restores the individual tree data, and makes up for the difficulty of not being able to monitor the rubber forest from the individual tree data. Through the present invention, the growth of each rubber tree can be monitored in real time, and the rubber trees can be picked in time, which can effectively avoid the product losses caused by natural disasters.

[0112] More importantly, the present invention can reasonably divide the vertical layered content of a single tree through a bottom-up single tree segmentation method. When forming single tree point cloud data, the amount of calculation can be minimized while ensuring accuracy. Under the same computing power resources, the obtained single tree point cloud data is more consistent with the actual growth of the single tree. Through the present invention, the prerequisite is provided for effectively monitoring the rubber forest and truly restoring the growth of the rubber forest.

[0113] In this embodiment, the variation law of the point cloud density in the vertical structure direction of the forest is that the point cloud density of the understory vegetation layer is high, the distribution is not concentrated, and the distribution is wide; the point cloud density of the trunk layer is the lowest, the distribution is concentrated, and the range is the smallest; the point cloud density of the canopy layer is relatively high, the distribution is concentrated, and the distribution range is between the understory vegetation layer and the trunk layer. The understory screening strategy in this embodiment is to set the minimum plant height of a single rubber tree, and delete the point cloud below the minimum plant height as shrub point cloud data.

[0114] The present invention also provides a segmentation system based on the above segmentation method. Through the system, the three-dimensional point cloud data of the rubber forest that has been collected can be quickly segmented into single trees through the above segmentation method on any intelligent terminal, and the growth conditions of each rubber tree can be restored more realistically, providing a prerequisite for accurate monitoring of the entire rubber forest. The rubber trees can be picked in time according to their actual growth conditions, and the yield loss caused by natural disasters can be effectively avoided.

[0115] In this embodiment, the UAV laser radar is used to collect data from the rubber forest in the full leaf stage to form the initial data of the full leaf stage, and collect data from the rubber forest in the deciduous stage to form the initial data of the deciduous stage; the initial data of the full leaf stage and the initial data of the deciduous stage are preprocessed respectively to obtain the point cloud data of the full leaf stage and the point cloud data of the deciduous stage; the point cloud data of the two stages are divided into blocks according to the position to form multiple groups of rubber forest point cloud data; each group of rubber forest point cloud data in the deciduous stage is vertically layered from bottom to top, and each group of point cloud data in the deciduous stage is divided into The data is divided into an understory vegetation layer, a trunk layer and a canopy layer, and the understory low vegetation is removed; afterwards, the basic skeleton point cloud data of the rubber tree in the deciduous period is extracted by a hierarchical clustering method to remove the point cloud data of the understory vegetation layer as the calculation data for the initial single tree segmentation, and finally the single tree segmentation of the rubber tree in the deciduous period is realized by a bottom-up single tree segmentation method; the point cloud data of the full leaf period is based on the single tree segmentation of the deciduous period, and the single tree segmentation of the rubber tree in the full leaf period is realized by taking the basic skeleton point cloud of the single tree in the deciduous period as the initial calculation data. The single tree point cloud data formed by the present invention is more accurate.

[0116] Embodiment 2

[0117] Different from the first embodiment, in this embodiment, each set of single tree point cloud data in the deciduous period is vertically layered from bottom to top, including the following contents:

[0118] Firstly, for each set of single tree point cloud data in the deciduous period, vertical stratification is performed from bottom to top according to the vertical distribution curve of point cloud density to obtain the understory vegetation layer data and the overstory data in the deciduous period.

[0119] Then, the hierarchical DBSCAN clustering algorithm was used to extract the basic skeleton points of rubber trees in the deciduous period, and the improved bottom-up region growing algorithm was used to segment the deciduous forest data based on the basic skeleton points to obtain the corresponding deciduous trunk layer data and deciduous canopy layer data.

[0120] First, distinguish the understory vegetation layer data during the deciduous period and the above-story data during the deciduous period, so that the understory vegetation layer data that does not need to participate in subsequent calculations can be quickly screened out. For the above-story data during the deciduous period, the geometric shape of the rubber tree is fully considered. The basic skeleton points are first determined, and then the trunk layer and canopy layer of the corresponding shape are confirmed based on the basic skeleton points. The entire calculation method is both rough and fine, and can maximize the calculation accuracy of the trunk layer and canopy layer under the premise of limited calculation amount, while effectively saving calculation steps.

[0121] Embodiment 3

[0122] In the present embodiment, more accurately full leaf period and leaf falling period are marked, it is convenient to automatically obtain corresponding data. For the time period below 15 ° and lasting more than one month, it is defined as leaf falling period, and the rest are full leaf period. For example, the corresponding full leaf period of rubber forest is the time period of September-October, and the corresponding leaf falling period is the time period of February of the following year. The three-dimensional point cloud data of rubber forest is collected in the time period of September-October every year to form the full leaf period initial data, and the three-dimensional point cloud data of rubber forest is collected in the time period of February of the following year to form the leaf falling period initial data. In the present embodiment, full leaf period and leaf falling period are found through research, and the more obvious growth characteristics of rubber tree are related to the accuracy of full leaf period and leaf falling period definition, which is directly related to the accuracy and true reproducibility of the single wood point cloud data formed.

[0123] Embodiment 4

[0124] In this embodiment, segmenting the leaf-falling period point cloud data into multiple groups of leaf-falling period single tree point cloud data includes the following steps:

[0125] First, through cluster analysis, we first find the center point of each tree in the leaf-falling point cloud data collected from the entire rubber forest;

[0126] Then, taking the center point as the center of the circle and the preset length as the radius, multiple groups of single tree point cloud data in the deciduous period are formed.

[0127] When looking for the center point, the point with the most concentrated cluster analysis or the point with the highest density is the center point of each set of deciduous point cloud data, and the preset length set in this scheme is 0.8-1.5 times the average radius of the recently announced adult rubber trees. Using the above method, multiple sets of deciduous single tree point cloud data can be quickly segmented, and each set of deciduous single tree point cloud data corresponds to a deciduous rubber tree.

[0128] In this embodiment, segmenting the point cloud data at the peak leaf stage into multiple groups of single tree point cloud data at the peak leaf stage includes the following steps:

[0129] First, the center point of each set of single tree point cloud data in the deciduous period is taken as the center of the circle;

[0130] Then, the point cloud data of the full leaf stage are divided into multiple groups of single tree point cloud data of the full leaf stage according to the preset length which is the same as the single tree point cloud data of the deciduous stage as the radius.

[0131] In the peak leaf period, due to the luxuriant branches and leaves and large crown width, it is very easy for two trees to form adhesions. The formation of single tree point cloud data in the deciduous period provides a basis for the division of single tree point cloud data in the peak leaf period. Because a tree is in a continuous and complete growth cycle in an adjacent peak leaf period and deciduous period, the trunk position of the same rubber tree will not change, and the chest radius range of the trunk will not change too much. Therefore, using the trunk data of the deciduous period for calculation not only does not affect the accuracy, but also saves calculation steps.

[0132] Through this embodiment, another point cloud single tree segmentation method can be provided, which can complete the remote monitoring of the entire rubber forest and each rubber tree on the basis of obtaining single tree data, so as to more accurately reflect the actual growth situation.

[0133] Embodiment 5

[0134] In this embodiment, when the leaf-falling period block data is segmented using the bottom-up improved region growing algorithm based on the basic skeleton points, Figure 3 As shown, this embodiment is based on the characteristics that the trunk of the rubber tree is straight and the understory vegetation is mostly low shrubs, combined with the bottom-up growth characteristics of the trees, the point cloud data collected in the deciduous period can clearly express the shape of the rubber tree poles, and the improved regional growing algorithm is used to segment the deciduous single tree, and the specific steps are as follows:

[0135] In the first step, all skeleton points in the first layer are used as seed points of individual trees, sorted according to the height of the point cloud, and starting from the lowest point, they become the initial seed points of the target tree to perform single tree segmentation. Because there are still a small number of short trees in the rubber forest, the distance between these short trees and rubber trees is not fixed. In order to more accurately segment short trees and rubber trees, combined with the straight trunk characteristics of rubber trees, the skeleton points of the first layer from bottom to top, that is, the bottom layer, are used as the seed points of each tree for initial seed point setting. For points other than the first layer, the initial seed point is determined according to the horizontal distance threshold. At the same time, in order to avoid branches being mistakenly divided into individual trees, the highest search threshold H of the seed point is set. seed , only points below the highest search threshold are judged as the initial seed points of a single tree.

[0136] In the second step, the skeleton points of individual trees are identified from the initial seed point upwards, the skeleton points belonging to the same tree are classified into one category, and other tree points are excluded according to the distance between the two points. That is, according to the planting rules of rubber trees and the size of the crown, the horizontal distance threshold T is set. HD , points with a distance greater than the horizontal distance threshold are removed from the target tree, and points within the horizontal distance threshold are classified according to the minimum distance principle.

[0137] The third step is to set a minimum plant height threshold H to prevent low shrubs from being mistakenly classified as rubber trees.tree , excluding low shrubs whose individual height is less than the threshold.

[0138] like Figure 3 As shown in the figure, when the rubber tree skeleton points are segmented, the lowest point A is regarded as the seed point of the target tree 1. Starting from point A, it "grows" upward according to the growth law of trees. If the distance between point B and point A is greater than THD, and point B is not a first-layer point, point B is considered to be a point of tree 2; if point B is a point at the bottom layer, point B is directly considered to be a point of tree 2. Assuming that there are many points belonging to tree 1 between point A and point C (including C), the distance between point D and point C is the smallest and less than THD, so point D belongs to tree 1, and point E also belongs to tree 1. Although point F is closer to point G, the horizontal distance between point F and book 2 is greater than THD, so point F belongs to tree 1. For point H, the horizontal distance from the two trees is equal and equal to THD. Since it has the smallest distance from point G, it is classified as tree 2.

[0139] In this embodiment, the horizontal distance threshold T HD The minimum plant height threshold is 0.7-1.1 times the average radius of the crown during the peak leaf period. tree The actual collected height of the trunk in the deciduous period and the height of the trunk in the full leaf period is 0.8-1 times of the minimum value. This setting can minimize the amount of calculation under the premise of meeting a certain segmentation accuracy. This solution can quickly achieve single tree segmentation.

[0140] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A bottom-up point cloud single tree segmentation method, characterized in that: The following steps are involved: Step 1: using the UAV laser radar, collect the initial data of the rubber forest in the full leaf stage to form the full leaf stage, and collect the initial data of the rubber forest in the deciduous stage to form the deciduous stage; Step 2, preprocessing the initial data of the full leaf stage and the initial data of the leaf falling stage respectively to obtain the point cloud data of the full leaf stage and the point cloud data of the leaf falling stage; Step 3, dividing the point cloud data of the full leaf stage and the point cloud data of the leaf falling stage into blocks according to the positions to form multiple groups of block data of the full leaf stage and block data of the leaf falling stage; Step 4: vertically stratify each set of deciduous period block data from bottom to top, and divide each set of deciduous period block data into understory vegetation layer, trunk layer and canopy layer according to the variation law of point cloud density in the vertical structure direction of the forest, and remove the understory low vegetation data contained in the understory vegetation layer according to the preset understory screening strategy to obtain deciduous period block processing data; Step 5, extracting the basic skeleton point cloud data of the rubber tree in the leaf-falling stage by using a hierarchical clustering method from the leaf-falling stage block processing data as the leaf-falling stage skeleton data; Step 6, for the deciduous period skeleton data, a bottom-up segmentation method is used to implement the single tree segmentation of the deciduous period rubber tree, to obtain the deciduous period single tree trunk data and the deciduous period single tree crown data, and the deciduous period single tree trunk data and the deciduous period single tree crown data are combined to form the deciduous period single tree point cloud data; Step 7, using the deciduous period skeleton data as the full leaf period skeleton data of the full leaf period block data corresponding to the group of deciduous period block data, and dividing each group of full leaf period block data from bottom to top into full leaf period single tree trunk data and full leaf period single tree crown data in the same manner as step 6; and combining the full leaf period single tree trunk data and the full leaf period single tree crown data to form full leaf period single tree point cloud data; Step 8: Combine the single tree trunk data in the deciduous period with the single tree trunk data in the full leaf period to form the single tree trunk data, combine the single tree crown data in the deciduous period with the single tree crown data in the full leaf period to form the single tree crown data; and combine the single tree trunk data and the single tree crown data to form the single tree point cloud data.

2. The bottom-up point cloud single tree segmentation method according to claim 1, characterized in that: The peak leaf period refers to the period when the crown of the rubber tree grows vigorously and more than 80% of the new leaves grow into old leaves; the deciduous period refers to the period when the crown of the rubber tree withers and more than 80% of the leaves fall.

3. The bottom-up point cloud single tree segmentation method according to claim 1, characterized in that: In step 5, the basic skeleton point cloud data of the rubber tree in the deciduous period is extracted using a hierarchical clustering method from the deciduous period block processing data as the deciduous period skeleton data, which includes the following steps: Step 51, stratifying the point cloud, stratifying the leaf-falling period block-processed data from bottom to top according to a certain height; Step 52, using a clustering algorithm for each layer of point cloud, clustering the point cloud of the branches and trunks of the rubber tree to form multiple groups of point cloud clusters; Step 53, obtain the center point of the point cloud in each point cloud cluster as the basic skeleton point of the cluster, extract all the basic skeleton points in a set of deciduous period block processing data to form the deciduous period skeleton data of the block area, and extract all the basic skeleton points in all the block areas to form the deciduous period skeleton data of the forest area.

4. The bottom-up point cloud single tree segmentation method according to claim 3, characterized in that: In step 6, a bottom-up segmentation method is used to achieve single-tree segmentation of rubber trees in the deciduous period, including the following steps: Step 61, using the skeleton data of the leaf-falling period as the basic data, taking the lowest point of the point cloud height as the initial seed point, and starting to grow upward; Step 62, setting two horizontal distance thresholds according to the planting rules of the rubber forest, starting from the initial seed point, calculating the distance and horizontal distance between the point clouds, and classifying the points with the smallest distance between the two points and less than the horizontal distance threshold as the seed points of the same tree; Step 63, classifying all basic skeleton points to complete the extraction of the basic skeleton point cloud of a single tree in the entire rubber forest, and forming multiple groups of basic skeleton seed point sets; Step 64, classifying all the deciduous point cloud data according to the one-to-one correspondence of the clusters in step 52, so as to achieve the single tree segmentation of the deciduous rubber trees.

5. The bottom-up point cloud single tree segmentation method according to claim 1, characterized in that: In step 5, the hierarchical DBSCAN clustering algorithm is used to extract the basic skeleton points of rubber trees in the deciduous period.

6. The bottom-up point cloud single tree segmentation method according to claim 5, characterized in that: When using the hierarchical DBSCAN clustering algorithm to extract the basic skeleton points of rubber trees in the deciduous period, the axis of the trunk is defined to be consistent with the crown. The points on the axis are first determined, and then the positions of the lateral end points of the trunk and crown are determined based on the symmetry settings. Then, the positions of the edge points of the trunk are determined based on the breast radius range, and the positions of the edge points of the crown are determined based on the crown width to form basic skeleton points.

7. The bottom-up point cloud single tree segmentation method according to claim 6, characterized in that: When the improved bottom-up region growing algorithm is used to segment the leaf peak period block data based on the basic skeleton points, the breast radius range is dynamically adjusted, and when the position of the trunk edge point is asymmetric, it is corrected according to the actual point cloud position collected.

8. The bottom-up point cloud single tree segmentation method according to claim 1, characterized in that: In step 8, the single tree trunk data in the deciduous period and the single tree trunk data in the full leaf period are combined to form the single tree trunk data, including the following contents: The single tree trunk data in the deciduous period and the single tree trunk data in the full leaf period are overlapped, and the breast radius range of the deciduous period trunk and the full leaf period trunk are compared by comparing the position coordinates of the trunk edge points. If the breast radius range of the deciduous period trunk is greater than or equal to the breast radius range of the full leaf period trunk, the single tree trunk data in the deciduous period is regarded as the single tree trunk data; If the breast radius range of the trunk in the deciduous period is smaller than the breast radius range of the trunk in the full leaf period, the tree will be marked as diseased.

9. The bottom-up point cloud single tree segmentation method according to claim 1, characterized in that: In step 8, the single tree crown data in the deciduous period and the single tree crown data in the full leaf period are combined to form the single tree crown data, including the following contents: The crown data of a single tree in the deciduous period and the crown data of a single tree in the full leaf period are overlapped, and the crown width length ranges of the crowns in the deciduous period and the full leaf period are compared by comparing the position coordinates of the crown edge points. If the crown width length range of the crown in the full leaf period is larger than that of the crown in the deciduous period, the data set formed by shrinking the crown width length of the edge position points of the crown data of the single tree in the full leaf period by more than one third is used as the crown data of the single tree; If the crown width length range of the crown layer in the peak leaf period is less than or equal to the crown width length range of the trunk layer in the deciduous period, the tree will be marked as diseased.

10. A bottom-up point cloud single tree segmentation system, characterized in that: The bottom-up point cloud single tree segmentation method as claimed in claim 1 is adopted.

Images