A non-contact method for measuring the breast height diameter of a target tree based on the form of the tree

Abstract

The application provides a non-contact standing tree breast diameter measurement method, comprising: obtaining a two-dimensional image and a ranging value of a measuring device to a target standing tree surface as input data; identifying a trunk region in the two-dimensional image obtained in a natural scene, comprising: first, establishing a region of interest image, determining the position of the target standing tree in the image and creating a cropped region of interest image therefrom; then, processing the region of interest image based on a trunk region identification model trained from a self-constructed dataset to identify the trunk region; detecting the trunk inclination and calculating the breast diameter length in the two-dimensional image, establishing a DBH calculation model, and calculating the DBH of the target standing tree. The application has a simple structure, can accurately and effectively extract the target trunk region in the image in various complex environments, takes into account the trunk shape feature and growth feature, and effectively improves the measurement accuracy of the DBH.

Description

TECHNICAL FIELD

[0001] The present application relates to a non-contact method for measuring the diameter at breast height of a standing tree, and more particularly to a method for measuring the diameter at breast height of a standing tree based on the shape of the target standing tree, belonging to the field of forestry structure parameter measurement. BACKGROUND

[0002] As an important parameter for evaluating the growth status of standing trees or forest stands, diameter at breast height (DBH) measurement has always been an important part of forest resource survey work. Implementing DBH measurement for each tree at the forest stand level can not only serve as an important input for estimating tree height, canopy height, and other structural parameters at the forest stand level, but also plays an important role in forest biomass estimation and revealing forest carbon cycling, carbon flow, and even global climate change.

[0003] DBH is defined as the diameter of the tree trunk at 1.3 m from the ground. Traditional DBH measurement methods generally use calipers and DBH tapes, which have high measurement accuracy and are often considered as true values. However, these tools are time-consuming and labor-intensive, and depend on the observation level of forestry technicians. Unlike the traditional contact measurement methods described above, the non-contact measurement methods based on remote sensing sensors developed in recent years have greatly enriched the means of DBH measurement. Their automated data processing mode and high measurement accuracy have made them applicable in some forest resource survey tasks.

[0004] In the non-contact method based on point cloud data to reconstruct the three-dimensional profile of the tree trunk and estimate the structure parameters of the standing tree, according to the different ways of obtaining point cloud, it is divided into two modes based on active sensor and passive sensor. Passive sensors generally use monocular or binocular vision derived machine vision methods to collect multiple images in the target area according to the planned route, and use image matching and motion recovery reconstruction technology to generate high-density point cloud to realize three-dimensional scene description and parameter estimation of the target area. Although this data acquisition and processing mode can meet certain accuracy requirements and has low cost, the complex environmental conditions in natural forests have always restricted the application and expansion of this method. In addition, the motion recovery reconstruction technology in the data processing process is a time-consuming task and requires certain computer performance, which further limits the effective application of this method.

[0005] Compared with the data acquisition method using passive sensors and based on machine vision technology, the data acquisition method based on discrete LiDAR (Light Detection and Ranging) derived ground laser scanning or mobile laser scanning can directly and actively obtain three-dimensional point cloud information of the target area and perform vegetation structure parameter inversion. With the development of current LiDAR data acquisition and point cloud data processing algorithms, the use of discrete point clouds to estimate the structure parameters of the target area has reached a high precision. However, the method based on LiDAR point cloud data to estimate the structure parameters still has certain shortcomings in data processing, hardware cost and portability. Although the current LiDAR data acquisition and processing method is relatively mature, the complex data processing method generates certain implicit costs; in addition, the LiDAR hardware cost is high and relatively bulky compared with traditional measurement tools, which further limits the application of LiDAR as a new technology in forest resource investigation.

[0006] Unlike the estimation of DBH based on three-dimensional point cloud data, there are related studies on the use of cameras to obtain two-dimensional images containing target trees and to non-contact measure DBH based on camera imaging principles. However, due to the development of digital cameras and image processing algorithms, this method of measuring DBH based on two-dimensional images still needs to be optimized and improved. First, the existing method of estimating DBH based on two-dimensional images relies on the segmentation and extraction of the trunk area in the image, although some studies have identified the trunk area by traditional threshold segmentation algorithm or manual selection. However, the complex and variable environmental conditions in natural forests limit the universality of traditional image segmentation algorithms, and manual identification of the trunk, although with high accuracy, undoubtedly reduces the measurement efficiency and puts forward more requirements for the professional quality of the measurers, so how to realize the automatic and strong adaptability of the trunk recognition algorithm still needs to be solved. Second, the use of two-dimensional images to realize non-contact measurement of DBH is generally based on optical imaging model and combined with depth information to establish spatial conversion relationship, while the existing DBH measurement model is only a simple application of optical imaging principle, and the form features of standing trees under natural conditions are not systematically considered, which reduces the applicability and measurement accuracy in natural forests. Therefore, a new DBH calculation model needs to be developed to take into account the influence of the form of the target standing tree on the measurement. SUMMARY

[0007] To solve the above problems, the purpose of the present application is to provide a method based on two-dimensional images and capable of non-contact DBH measurement under various forest stands. First, the method establishes a spatial conversion relationship from image space to real three-dimensional space using depth information obtained by a laser range finder. Second, the method can automatically identify the trunk area in the image under various complex scenes by combining the measurement features and using a related open-source deep learning framework. Finally, an accurate DBH calculation model is established based on the principle of optical imaging and combined with the shape features and inclination direction of the standing tree trunk, thereby improving the DBH measurement accuracy.

[0008] To solve the above problems, the present application provides the following technical solutions:

[0009] A non-contact standing tree DBH measurement method, characterized in that it comprises the following steps:

[0010] Step one: obtaining two-dimensional images and distance measurement values from the target standing tree surface as input data; Step two: identifying the trunk area in the two-dimensional images obtained in the natural scene, including: first, establishing a region of interest image, determining the position of the target standing tree in the image and creating a cropped region of interest image; then processing the region of interest image based on the trunk area recognition model trained by the self-constructed dataset to identify the trunk area; Step three: trunk inclination detection and breast diameter length L pixel calculation in the two-dimensional image pixel as shown in the following formula:

[0011]

[0012] wherein d left and d right are the distances from the center of gravity of the trunk area to the two fitted straight lines of the trunk edge contour, and alpha and beta are the inclination angles of the two fitted straight lines of the trunk edge contour in the two-dimensional image relative to the horizontal line;

[0013] Step four: establishing a DBH calculation model to calculate the DBH of the target standing tree, including: establishing an optical imaging model as shown in the following formula:

[0014]

[0015] wherein the included angle formed by the two intersection points of the imaging light and the trunk surface is respectively set as A, B and 2alpha, and f is the focal length of the measurement device; d l is the distance from the measurement device to the intersection line AB; DBH l is the length of the intersection line AB; a DBH calculation model is established to calculate the DBH of the target standing tree as shown in the following formula:

[0016]

[0017] Wherein, the target tree center is represented as O, d is the distance from the measuring device to the target tree surface, d c is the distance from the measuring device to the target tree center O.

[0018] Further, in step one, the measuring device is a device integrated with a camera and a laser range finder, a two-dimensional image is obtained through the camera, and a ranging value is obtained through the laser range finder.

[0019] Further, in step two, the position of the target tree in the image is determined by positioning the light spot formed on the trunk when the laser range finder is working, and the cropped image of the region of interest is created by cropping the original image with a height of 200 pixels and a width of n pixels as the center of the light spot coordinates.

[0020] Further, in step two, the adaptability of the trained model to different scenes is ensured by considering various environmental factors in natural scenes to construct a training data set as input, and the recognition model is trained based on the deep learning U-Net network.

[0021] Further, the light spot formed on the trunk surface when the laser range finder is working is used as a recognition feature to locate the position of the target tree in the image, including: first, converting the original image from RGB space to HSV space, outputting a binary image by setting an empirical global threshold, and extracting all potential light spot pixels; then calculating the first moment of each pixel block to obtain the corresponding barycenter coordinates, representing the position of the pixel block in the image; assuming that the barycenter of the light spot is distributed in the center region of the image and is relatively isolated from other barycenter points, then a normalized expression for extracting the barycenter of the light spot is established:

[0022]

[0023] Wherein, is the number of all potential barycenter points, which is calculated by counting the total number of barycenter points in the image; n p is the number of other barycenter points close to the pth barycenter point after setting the search radius, which is used to measure the isolation degree of the current barycenter point, and is determined by counting the number of barycenter points within the search radius of the pth barycenter point; m is a global weight factor, set to 100; d image is the distance from the image center to the farthest pixel, which is a constant value after the camera is determined, i.e. the pixel distance from the image center to the four corner points of the image; d p is the distance from the pth barycenter point to the image center, which is obtained by calculating the Euclidean distance between the barycenter point pixel coordinates and the image center; K p then represents the probability that the pth barycenter point is a light spot, and when K p is the smallest, point p is considered as the target tree position, and when the calculation of formula 1 is completed, the target tree position is obtained.

[0024] Further, the trunk inclination detection in step three includes binarizing the trunk region identified in step two, and calculating the region center of gravity, the outer bounding box, and the edge contour, respectively.

[0025] Further, the region center of gravity is obtained by calculating the second moment of the binarized region of the image; the outer bounding box is obtained by determining the maximum circumscribed rectangle through counting the edge coordinates around the binarized region; the edge contour is obtained by traversing each pixel in the binarized image row by row to determine the left and right edge pixel positions of each row, and then performing linear fitting on the left and right edge contours using the least squares method to obtain the two side inclination angles a and β, and the slopes of the two fitted straight lines are converted into angle values a and β to represent the inclination of the left and right edges of the trunk.

[0026] Further, the breast diameter length L pixel The calculation step includes, taking the trunk region center of gravity as the center, calculating the perpendicular lines to the two fitted straight lines and the lengths DBH L and DBH R , and the average of the two is the breast diameter length L pixel , wherein DBH L and DBH R are as follows:

[0027]

[0028] wherein d left and d right are the distances from the trunk region center of gravity to the two fitted straight lines of the trunk edge contour; a and β are the inclination angles of the two fitted straight lines of the trunk edge contour in the two-dimensional image relative to the horizontal line; DBH L and DBH R represent the pixel lengths of the breast diameter calculated from the two fitted straight lines.

[0029] Further, wherein formula 4 and formula 5 are solved jointly, and DBH is represented as follows:

[0030]

[0031] 10. The measurement method according to claim 9, wherein based on formula 6, DBH is obtained as follows:

[0032]

[0033] wherein,

[0034]

[0035] The application provides a non-contact DBH measurement method based on target tree form, which is based on two-dimensional images and considers the target tree form, is different from a traditional DBH calculation method based on a simple application of an optical imaging principle, and improves the DBH measurement precision of the idea applied to a complex natural forest by considering the characteristics of the shape of a trunk, the growth characteristics of a tree in a natural scene, and variable environmental factors. First, the method can automatically and accurately identify the target trunk in the image in various extreme environments, then a DBH mathematical calculation model is established by considering the actual imaging characteristics of the trunk and detecting the growth direction of the trunk. The method has the following two advantages: 1) the target trunk region in the image can be accurately and effectively extracted in various complex environments; and 2) the calculation model established by considering the shape characteristics and growth characteristics of the trunk can effectively improve the DBH measurement precision. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 is a flowchart of the measurement method of the application.

[0037] Figure 2 is a trunk tilt detection schematic diagram.

[0038] Figure 3 is a DBH calculation schematic diagram considering the shape of the trunk.

[0039] Figure 4 is a scatter plot and a box plot for comparing the results of the method (DBH Measurement Model, DMM) and the simple optical imaging model (Optical Imaging Model, OIM) of the application with the chest diameter girth measurement results, (a) scatter plots and linear fitting results of the two methods; (b) error box plots of the two methods. DETAILED DESCRIPTION

[0040] The technical solutions of the application are described in detail below with reference to the drawings.

[0041] The application provides a non-contact tree DBH measurement method based on target tree form, in particular, a calculation method for non-contact DBH measurement based on two-dimensional images and considering the target tree form. First, the method obtains two-dimensional images and depth information as input; second, the automatic identification of the trunk in various scenes is realized by means of image processing and depth learning algorithm; finally, the shape and growth characteristics of the tree are fused with the basic optical imaging model, so as to improve the DBH calculation precision. The following takes the specific DBH calculation process as an example for description, and the flowchart is shown in Figure 1 The specific implementation steps are as follows:

[0042] Step one: obtaining input data. Specifically, obtaining a two-dimensional image and a ranging value of the measuring device to the target tree surface as input data, preferably, the measuring device is an integrated device of camera and laser range finder, by fixing the two devices on a connecting fixture, measuring the distance between the camera optical center and the ranging center at the same time, thus obtaining the integrated device, and starting both devices at the same time after aiming at the target tree to obtain the two-dimensional image and the ranging value of the integrated device to the target tree surface, but the present application does not make specific limitation to the measuring device. Among them, the ranging value is regarded as the depth information of the target tree in the two-dimensional image, and the two-dimensional image containing the target tree and the depth information to the target tree surface are obtained by camera and laser range finder respectively, preferably, the camera can be a special digital camera, an optical camera or a camera of a smart terminal, but the present application does not make specific limitation to this. See Figure 1 , wherein the depth information is input into the DBH calculation model established on the basis of the optical imaging model to realize the conversion of the two-dimensional image to the real three-dimensional space; the two-dimensional image is subsequently used to extract the trunk region and calculate the pixel size of DBH, and finally input into the established DBH calculation model.

[0043] Step two: identifying the trunk region in the two-dimensional image obtained in the natural scene by image processing algorithm.

[0044] See Figure 1 , which can be summarized as the following two steps: 1) establishing a region of interest (ROI) image, determining the position of the target tree in the image and creating a cropped region of interest (ROI) image therefrom. Among them, the position of the trunk in the image is determined by positioning the light spot formed on the target trunk when the laser range finder is working, and the original image (m (high) x n (wide) pixels) is cropped to an ROI image with a height of 200 pixels and a width of n pixels with the light spot coordinates as the center, so as to enhance the saliency of the trunk in the image, improve the identification accuracy of the trunk in the subsequent step, and at the same time, compress the size of the image to be processed and improve the calculation processing efficiency; 2) processing the ROI image based on the trunk region recognition model trained by the self-constructed data set to recognize the trunk region. In order to ensure that the recognition model has strong universality and robustness, the training data set is self-constructed by considering various environmental factors in the natural scene as input to ensure the adaptability of the trained model to different scenes. In addition, the recognition model is trained based on the conventional deep learning U-Net network, and finally a trunk recognition model with good universality and high robustness is obtained.

[0045] For the two-dimensional image of multiple trees, the laser rangefinder forms a light spot on the surface of the tree trunk when it is working, which is used as a recognition feature to locate the position of the target tree in the image. First, the original image is converted from RGB space to HSV space, and a binary image is output by setting an empirical global threshold to extract all potential pixel blocks as light spots. Then, the first moment of each pixel block is calculated to obtain the corresponding barycentric coordinates, representing the position of the pixel block in the image. By taking advantage of the short distance between the camera optical center and the ranging center when the device is integrated, and assuming that the barycentric coordinates of the light spot are distributed in the center area of the image and are relatively isolated from other barycentric coordinates, a normalized expression for extracting the barycentric coordinates of the light spot can be established:

[0046]

[0047] where, is the number of all potential barycentric points, which is calculated by counting the total number of barycentric points in the image; n p is the number of other barycentric points close to the pth barycentric point after setting the search radius, which is used to measure the degree of isolation of the current barycentric point, and is determined by counting the number of barycentric points within the search radius of the pth barycentric point; m is a global weight factor, which is an empirical parameter and is generally set to 100; d image is the distance from the image center to the farthest pixel, which is generally a constant value after the camera is determined, i.e., the pixel distance from the image center to the four corners of the image; d p is the distance from the pth barycentric point to the image center, which is obtained by calculating the Euclidean distance between the barycentric point pixel coordinates and the image center; K p represents the probability that the pth barycentric point is a light spot, and when K p is the smallest, the point p is considered as the target tree position. When the calculation of equation 1 is completed, the target tree trunk region can be obtained.

[0048] The extraction of the target tree trunk region is achieved by self-building a dataset and training a tree trunk region recognition model based on the open-source deep learning framework PaddlePaddle. The main deep learning image segmentation algorithm used is based on the U-Net network, which can effectively utilize labeled data from limited samples by first extracting features in the encoder and then upsampling in the decoder, achieving accurate segmentation and recognition of tree trunk regions in different scenarios. At the same time, the training set considers multiple environmental factors when establishing to ensure the robustness of the recognition model, including changes in lighting, measurement scenarios, tree species, and image shooting distance, etc., ultimately obtaining a tree trunk region recognition model with high robustness and strong universality.

[0049] This step is different from the traditional DBH solving method based on the simple application of optical imaging principle. By combining the optical imaging model with depth information, based on the tree trunk area recognition model trained by self-constructed dataset, the characteristics of tree trunk shape, natural scene growth characteristics and variable environmental factors are considered, which can accurately and effectively extract the target tree trunk area in the image under various complex environments, and can automatically and accurately identify the target tree trunk in the image under various extreme environments.

[0050] Step three: trunk inclination detection and diameter at breast height length calculation in two-dimensional image.

[0051] Based on the trunk area extracted in step two, the input of the trunk inclination detection is obtained. In order to accurately represent the inclination state of the target tree trunk in the two-dimensional image, it is assumed that the inclination of the two ends of the trunk has certain difference and the overall inclination of the ROI can represent the inclination state of the target tree.

[0052] As shown in Figure 2 , the left and right sides of the trunk have different inclination angles. The target tree trunk area identified in step two is binarized, and the pixel values of the trunk part and the pixel values outside the trunk part in the image are set to two different pixel values, respectively, so as to obtain the trunk area binary image. On the basis of the trunk area extracted in step two, the center of gravity, the outer bounding box and the edge contour of the trunk area are calculated, respectively. The center of gravity of the trunk area is obtained by calculating the second moment of the binarized region of the image; the outer bounding box is obtained by calculating the maximum circumscribed rectangle of the binarized region; and the edge contour is obtained by traversing each pixel in the binarized image to determine the left and right edge pixel positions of each row. Then, the least square method is used to linearly fit the left and right edge contours to obtain the inclination angles α and β, respectively. The slopes of the two fitted straight lines are converted into angle values (α, β) to represent the inclination of the left and right edges of the trunk; finally, the perpendicular lines of the two edge contour fitting lines are obtained with the center of gravity of the trunk area as the center, and the lengths DBH L and DBH R of the reverse extension to the other straight line are obtained, and the average value of the two is the pixel length L pixel of the DBH, i.e. the length of the DBH in the two-dimensional image. The pixel length of the DBH calculated by the trunk inclination can be represented by the following formula:

[0053]

[0054] where d left and d right are the distances from the center of gravity of the trunk area to the left and right fitting straight lines, respectively; α and β are the inclination angles of the left and right fitting straight lines with respect to the horizontal line; DBH L and DBHR These represent the DBH pixel lengths calculated using the left and right slopes, respectively; the final DBH pixel length L pixel The specific expression is obtained through DBH L and DBH R The average calculation yields:

[0055]

[0056] Step 4: Establish a DBH calculation model and calculate the DBH of the target standing tree.

[0057] A DBH computational model is established by considering the tree trunk shape characteristics and their imaging characteristics on the camera. (See also...) Figure 3 This demonstrates the geometric relationship when a laser rangefinder and camera aim at a target tree for measurement. Based on a fundamental optical imaging model, the DBH calculation expression is derived using the constructed imaging geometry. Finally, by inputting depth information and DBH pixel length, non-contact, precise measurement of DBH is achieved.

[0058] Specifically, after extracting the tree trunk region and detecting its tilt in the 2D image, the physical length (in cm) of the tree trunk region DBH is calculated using camera intrinsic parameters. Simultaneously, a DBH calculation model is established based on optical imaging principles and incorporating depth information from the tree trunk surface. For standing trees in natural environments, the DBH calculation model treats the tree trunk as a cylindrical geometry and uses the depth value d from the camera to the center of the trunk. c This is expressed as the sum of the laser ranging value d and half of DBH; simultaneously, the length of the line connecting the intersection point of the camera's imaging ray and the tree trunk surface is not equal to DBH, and the magnitude of this difference is positively correlated with the measured DBH value of the standing tree. Therefore, the basic camera optical imaging model relationship should be written here as follows:

[0059]

[0060] Where f is the camera focal length; d l DBH is the distance from the camera's optical center to the intersection line AB. l L is the length of the intersection line AB; pixel The length of the DBH pixels obtained in step three is the length of the tree trunk region in the two-dimensional image. To quantitatively represent the above geometric relationships, the center of the tree trunk is represented as O, and the two intersection points of the imaging ray with the tree trunk surface and the included angle are respectively set as A, B, and 2α. The following combined solution formula can be derived based on the geometric relationship characteristics:

[0061]

[0062] After solving and simplifying equations 4 and 5 together, DBH can be expressed as the following cubic equation in one variable:

[0063]

[0064] The monic cubic equation listed in formula 6 is solved by setting a discriminant. Since the discriminant is always less than 0, the final expression of DBH can be fixed and solved as:

[0065]

[0066] In the formula,

[0067]

[0068] This step can effectively improve the measurement accuracy of DBH by considering the actual imaging characteristics of the tree trunk and detecting the growth direction of the tree trunk, taking into account the shape characteristics and growth characteristics of the tree trunk to establish a DBH mathematical solution model.

[0069] Specific measurement examples:

[0070] The trunk recognition model is implemented using the EasyDL module of the open-source deep learning framework. By collecting samples in various field conditions, a training and verification dataset is constructed, and a trunk recognition model with high recognition accuracy is trained. The dataset is mainly constructed from the perspectives of complex lighting conditions, measurement scenes, multiple tree species, and different depth values. The dataset ultimately consists of 572 images and is used to construct the trunk recognition model, of which 10% is used to verify the recognition accuracy of the recognition model. Among the three types of accuracy indicators commonly used in the field of deep learning, the precision is 92.7%, the recall is 100%, and the F1 score is 0.96, indicating that the recognition model can accurately identify the trunk area and can be applied in various measurement conditions.

[0071] The measurement algorithm was tested under three different forest conditions and compared with the algorithm based only on the basic optical imaging model. At the same time, the accuracy of the measurement was verified by the tree caliper. The three experimental forest conditions are mountain natural forest, artificial shelter forest, and urban artificial forest, which represent different degrees of complexity of background conditions. A total of 371 standing trees were measured, and the results were compared with the measurement results of the tree caliper. For example, Figure 4 ​(a) The figure shows the measurement results of the application (DMM) and the measurement results of the simple optical imaging model (OIM). It can be seen from the figure that the algorithm of the application has good consistency with the reference value of DBH obtained by tree girth measurement (horizontal axis in the figure), the fitting slope is close to 1, and the root mean square error RMSE is 1.40 cm. Compared with the method of simple calculation using the basic optical imaging model, the fitting straight line of the measurement result of the application is closer to the 1:1 line, indicating that it has higher accuracy. In addition, Figure 4 (b) The error box plot of the two algorithms in different DBH size intervals is shown. It can be seen that as the diameter of the standing tree increases, the method of simple calculation using the basic optical imaging model will produce obvious underestimation error of DBH, while the algorithm can effectively correct this influence. At the same time, the measurement results of OIM represented by the box plot have large upper and lower limits of the box in all diameter intervals, showing more uncertainty compared with DMM. Therefore, the DBH calculation method of the application can accurately measure the standing tree with a diameter of 5-55 cm in a non-contact manner, and compared with the traditional OIM calculation method, the DMM calculation method proposed in the application can significantly improve the measurement accuracy of the diameter of the standing tree.

Claims

1. A non-contact standing tree DBH measurement method, characterized by, The method comprises the following steps: Step one: obtaining a two-dimensional image and a ranging value of a measuring device to a target standing tree surface as input data; Step two: identifying a tree trunk region in the two-dimensional image obtained in a natural scene, comprising: first, establishing a region of interest image, determining the position of the target standing tree in the image and creating a cropped region of interest image; then, processing the region of interest image based on a tree trunk region identification model trained by a self-constructed data set to identify the tree trunk region; Step three: trunk inclination detection and breast height length L in two-dimensional image pixel L = 2 * sqrt (A / π) pixel As shown in the following formula: where d left and d right are the distances from the center of gravity of the trunk region to the two fitted straight lines of the trunk edge contour, and a and β are the tilt angles of the two fitted straight lines of the trunk edge contour relative to the horizontal line in the two-dimensional image, respectively. Step four: establishing a DBH calculation model to calculate the DBH of the target standing tree, comprising: establishing an optical imaging model, as follows: Wherein, the two intersection points of the imaging light and the trunk surface and the included angle formed are respectively set as A, B and 2α, f is the focal length of the measuring device; d l is the distance from the measuring device to the intersection line AB; DBH l is the length of the intersection line AB; a DBH calculation model is established to calculate the DBH of the target tree, as follows: where the center of the target tree is represented as O, d is the distance from the measuring device to the surface of the target tree, d c is the distance from the measuring device to the center of the target tree O. In step three, the trunk inclination detection comprises: performing binaryzation processing on the tree trunk region identified in step two, and calculating the region center of gravity, the outer bounding box and the edge contour, respectively; In step three, the trunk inclination detection comprises: performing binaryzation processing on the tree trunk region identified in step two, and calculating the region center of gravity, the outer bounding box and the edge contour, respectively; wherein the breast height length L pixel The step of calculating includes, with the trunk region barycenter as the center, respectively calculating the length DBH L and DBH R , and the average of the two is the breast height length L pixel wherein DBH L and DBH R are as follows: where d left and d right are the distances from the center of gravity of the stem region to the two fitted straight lines of the stem edge contour; a and b are the angles of inclination of the two fitted straight lines of the stem edge contour with respect to the horizontal line in the two-dimensional image; DBH L and DBH R represent the pixel lengths of the breast height diameters calculated from the two fitted straight lines, respectively.

2. The measurement method according to claim 1, wherein, In step one, the measuring device is a device integrating a camera and a laser range finder, which obtains a two-dimensional image through the camera and a ranging value through the laser range finder.

3. The measurement method according to claim 2, wherein, In step two, the position of the target standing tree in the image is determined by the light spot formed on the trunk when the laser range finder is working, and the cropped region of interest image is created by cropping the original image with a height of 200 pixels and a width of n pixels from the center of the light spot coordinates.

4. The measurement method according to claim 2, wherein, In step two, the adaptability of the trained model to different scenes is ensured by considering various environmental factors in the natural scene to self-construct a training data set as input, and the identification model is realized based on a deep learning U-Net network training.

5. The measurement method according to claim 3, wherein, The light spot formed on the trunk surface when the laser range finder is working is used as a recognition feature to locate the position of the target standing tree in the image, comprising: first, converting the original image from RGB space to HSV space, outputting a binary image by setting an empirical global threshold, and extracting all potential light spot pixels; then, calculating the first moment of each pixel block to obtain the corresponding center of gravity coordinates, representing the position of the pixel block in the image; assuming that the light spot center of gravity is distributed in the center area of the image and is relatively isolated from other center of gravity points, a normalized expression for extracting the light spot center of gravity is established as follows: wherein, is the number of all potential pixel block barycentric points, which is calculated by counting the total number of barycentric points in the image; n p is the number of other barycentric points close to the pth barycentric point after setting the search radius, which is used to measure the degree of isolation of the current barycentric point, and is determined by counting the number of barycentric points within the search radius of the pth barycentric point; n is a global weight factor, which is set to 100; d image is the farthest pixel distance to the image center, which is a constant value after the camera is determined, i.e., the pixel distance from the image center to the four corner points of the image; d p is the distance from the pth barycentric point to the image center, which is obtained by calculating the Euclidean distance from the barycentric point pixel coordinates to the image center; K p then indicates the probability that the pth barycentric point is a light spot, and when K p is the smallest, the point p is regarded as the target tree position, and when the calculation of formula 1 is completed, the target tree position can be obtained.

6. The measurement method of claim 1, wherein, Solving equations 4 and 5 jointly, the DBH is expressed as follows: 。 7. The measurement method according to claim 6, wherein, Based on equation 6, the DBH is obtained as follows: wherein,

Images