Image recognition-based forest tree annual growth data telemetry method and multispectral image recognition device

Abstract

The application relates to a forest tree annual growth data remote measurement method based on image recognition, which comprises the following steps: a multispectral image recognition device is controlled by a positioning algorithm to be aligned to a measured tree, and the current coordinates of the measured tree are recorded; the images of multiple spectrums of the measured tree are acquired; the multiple spectrums are recorded at the same time and have time point information; pixel cutting is performed on the multispectral images, the image edge feature mismatch area is removed, and a multispectral image set with completely coincided features is obtained; image normalization calculation is performed, and a normalized image is regenerated according to the normalization calculation result; the pixel numbers of tree height H, crown width A and diameter at breast height C are acquired, and the actual values of the tree height H, the crown width A and the diameter at breast height C are calculated; the RGB feature information of tree leaves is acquired, the specific growth cycle T of the measured tree is determined, and the annual growth data of the measured tree is acquired to obtain a growth model Phi of the measured tree; and data generation comprises the tree height H, the crown width A, the diameter at breast height C and the growth cycle T.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of forest data measurement, in particular to a forest annual growth data remote measurement method based on image recognition. BACKGROUND

[0002] Forests, as important natural resources in China, have very important significance for human survival and development. Effective forest resource information collection can provide scientific basis for forestry enterprises and institutions to formulate forest management plans, and also provide necessary basis for the state and governments at all levels to formulate economic and environmental plans and policies, which is conducive to providing scientific and effective management of forest resources.

[0003] The traditional forest growth data collection method mainly relies on manual measurement, which has problems such as low measurement efficiency, low measurement accuracy, and discontinuous data.

[0004] With the development of computer vision technology, image recognition technology has been gradually applied to the forestry field, and how to better collect forest data and build models has become a technical problem to be solved in the field.

[0005] For example, the prior art discloses a vegetation measurement method based on hyperspectral remote sensing and photogrammetry technology. The method uses spectral remote sensing technology and camera measurement technology to measure vegetation. It can measure tree height and diameter at any point of the trunk, tree crown projection area and crown width, tree crown surface area, and tree crown volume through mathematical models. However, this method is complex and cannot simply and effectively obtain the annual growth data of trees through the tree growth model. SUMMARY

[0006] In order to at least partially solve the above technical problems, the present application provides a forest annual growth data remote measurement method based on image recognition.

[0007] The forest annual growth data remote measurement method based on image recognition provided by the present application adopts the following technical scheme.

[0008] A forest annual growth data remote measurement method based on image recognition comprises:

[0009] The multispectral image recognition device is aligned to the measured tree under the control of the positioning algorithm, and the current coordinates of the measured tree are recorded;

[0010] Obtain images of multiple spectra of the measured tree; the multiple spectra are recorded and saved with time point information at the same time;

[0011] Pixel cutting is performed on the multispectral images to remove the non-matching area of the image edge features, and a multispectral image set with completely coincident features is obtained.

[0012] The image is normalized, and a normalized image is regenerated according to the normalization result;

[0013] The pixel numbers of the tree height H, the crown width A and the diameter at breast height C are obtained, and the actual values of the tree height H, the crown width A and the diameter at breast height C are calculated;

[0014] The RGB feature information of the leaves is obtained, and the specific growth period T of the measured tree is determined; and

[0015] The growth model Φ of the measured tree is fitted by using the growth data of the measured tree in a whole year; the growth data includes the tree height H, the crown width A, the diameter at breast height C and the growth period T.

[0016] Optionally, the method for obtaining the pixel numbers of the tree height H, the crown width A and the diameter at breast height C comprises:

[0017] Seven key points and a center line L are selected; the first key point o1 is the center point of the root of the measured tree; the center line L is a straight line passing through o1 and perpendicular to the ground; the second key point o2 is the top point of the contour of the measured tree; the third key point o3 is the left edge point of the contour of the measured tree farthest from the center line; the fourth key point o4 is the right edge point of the contour of the measured tree farthest from the center line; the fifth key point o5 is a point on the center line at a fixed distance from o1; the sixth key point o6 is the left contour point farthest from the center line at the same height as o5; and the seventh key point o7 is the right contour point farthest from the center line at the same height as o5.

[0018] In the formula, m represents the pixel number of the tree height H, n represents the pixel number of the crown width A, and t represents the pixel number of the diameter at breast height C.

[0019] The pixel number m of the tree height H is o2-o1.

[0020] The pixel number n of the crown width A is o4-o3.

[0021] The pixel number t of the diameter at breast height C is o7-o6.

[0022] Optionally, the method for obtaining the pixel numbers of the tree height H, the crown width A and the diameter at breast height C and calculating the actual values of the tree height H, the crown width A and the diameter at breast height C comprises:

[0023] The normalized image with the clearest contour and the highest contrast is selected;

[0024] The cluster centers are randomly selected, and all sample points are classified according to the current cluster centers;

[0025] The mean value of the sample points of each class in the current iteration is calculated as the cluster center of the next iteration;

[0026] The difference between the cluster center of the next iteration and the current cluster center is calculated;

[0027] If the gap is less than a given iteration threshold, the iteration ends, otherwise the iteration continues; the target of the cluster segmentation is arranged in a sequence from top to bottom and from left to right to form the contour of the measured tree;

[0028] According to the correspondence between the pixel size and the actual size, the tree height H, the crown width A, the breast diameter C and the actual values can be obtained.

[0029] Meanwhile, the application further provides a multispectral image recognition device, which comprises a visual camera and a holder;

[0030] The holder comprises a posture reading assembly, a stepping motor and a digital processing unit; the digital processing unit comprises a posture control module and a spectrum analysis module;

[0031] The visual camera outputs optical images to the digital processing unit, and different spectrum filters inside the visual camera are used to output images of different spectrums;

[0032] The posture reading assembly outputs the vertical rotation angle and the horizontal rotation angle of the holder to the digital processing unit;

[0033] The stepping motor comprises a vertical stepping motor and a horizontal stepping motor to drive the holder to rotate vertically and horizontally;

[0034] The posture control module sends instructions to the stepping motor to make the stepping motor rotate in at least one of the vertical angle and the horizontal angle;

[0035] The spectrum analysis module is used to process spectrum data at different vertical angles and horizontal angles and splice the spectrum data into panoramic spectrum images of the measured scene.

[0036] Optionally, the visual camera comprises a lens and a camera, the camera is fixedly connected with the lens, and the spectrum filter is installed inside the lens.

[0037] Optionally, the holder comprises a base for mounting the visual camera, one end of the base is provided with a viewing port, the lens is aligned with the viewing port, and the camera acquires external images and optical information from the viewing port through the lens.

[0038] Optionally, the holder further comprises a horizontally placed base and a vertical shaft rotatably mounted on the base at the lower end; a rotating platform is fixedly sleeved on the outer circumference of the vertical shaft; a vertical shaft encoder is arranged at the upper end of the vertical shaft, the vertical shaft encoder detects and outputs the horizontal rotation angle of the visual camera; and the vertical shaft is connected with the vertical stepping motor, and the vertical stepping motor drives the vertical shaft to rotate.

[0039] Optionally, a lower frame is fixedly installed on the rotating platform; the lower frame is in inverted U shape; the first horizontal shaft and the second horizontal shaft are installed on two sides of the lower frame and can rotate on the lower frame;

[0040] The two ends of the first horizontal shaft and the second horizontal shaft are fixedly connected with two sides of an inverted upper frame respectively;

[0041] The base is fixedly installed on the upper surface of the upper frame;

[0042] One end of the first horizontal shaft is provided with a horizontal shaft code disc, and one end of the second horizontal shaft is connected with a horizontal stepping motor; the horizontal stepping motor drives the first horizontal shaft and the second horizontal shaft to rotate;

[0043] The horizontal shaft code disc detects and outputs the vertical rotation angle of the visual camera.

[0044] Optionally, the posture control module sends instructions to the stepping motor, so that the stepping motor rotates appropriate horizontal and vertical angles, thereby driving the corresponding horizontal shaft and vertical shaft to rotate to the required horizontal and vertical angles;

[0045] The spectrum analysis module processes the spectrum data of the multispectral visual camera matrix at different vertical and horizontal angles to generate a panoramic spectrum image of the measured scene;

[0046] The spectrum analysis module pre-stores image stitching algorithms, frame synchronization algorithms and multispectral image processing algorithms. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is a method flowchart of the present application;

[0048] Figure 2 is a tree shape contour extracted from a normalized image by a clustering algorithm;

[0049] Figure 3 is a perspective view of a multispectral image recognition device of the present application;

[0050] Figure 4 is a perspective view of the multispectral image recognition device from another angle;

[0051] Figure 5 is a front view of the multispectral image recognition device of the present application;

[0052] In the figure, 1 is a lens; 2 is a camera; 3 is a filter; 4 is a base; 5 is an observation port; 6 is a base; 7 is a vertical shaft; 8 is a rotating platform; 9 is a vertical shaft code disc; 10 is a vertical stepping motor; 11 is a lower frame; 12 is a first horizontal shaft; 13 is a second horizontal shaft; 14 is an upper frame; 15 is a horizontal shaft code disc; and 16 is a horizontal stepping motor. DETAILED DESCRIPTION

[0053] The application will be further described below in conjunction with the accompanying drawings and specific embodiments: Figures 1-5

[0054] First of all, it needs to be explained here that in the description of the present application, if the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and other orientation words appear, the orientation or position relationship indicated thereby is based on the orientation or position relationship shown in the drawings, which is only for the convenience of description, and does not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application; in addition, if the terms "first", "second", "third" and the like appear, they are only for the purpose of description and cannot be understood as indicating or implying relative importance. In addition, in the present application, unless otherwise explicitly specified and limited, if the terms "mounting", "connecting", "connecting" appear, they should be understood in a broad sense, for example, it can be fixed connection, or detachable connection, interference fit, transition fit and other limit connection, or integral connection; it can be directly connected, or indirectly connected through an intermediate medium; therefore, for those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0055] The embodiments of the present application disclose a kind of forest all-year growth data telemetry method based on image recognition. Refer to Figure 1 As an embodiment of a kind of forest all-year growth data telemetry method based on image recognition, a kind of forest all-year growth data telemetry method based on image recognition includes the following steps:

[0056] Step 101, the multispectral image recognition device is aligned to the measured tree under the control of positioning algorithm, and the current coordinates of the measured tree are recorded.

[0057] Specifically, when using a multispectral image recognition device to measure a tree, the device needs to be aligned to the measured tree in order to obtain a multispectral image of the tree. In order to realize the positioning and recording of the measured tree, the multispectral image recognition device needs to be controlled using a positioning algorithm so that it can accurately align to the measured tree. The positioning algorithm can be implemented in various ways, such as using GPS positioning technology, image matching-based positioning algorithm, laser ranging-based positioning technology or wireless communication-based positioning technology, etc. The current coordinates of the measured tree can be obtained by the positioning algorithm, usually in the form of longitude and latitude. These coordinate information can be used for geographical location marking and monitoring of the tree.

[0058] Step 102, obtain the image of multiple spectra of the measured tree; multiple spectra are recorded at the same time and saved with time point information. ​

[0059] Specifically, when we want to understand the growth status, health condition and environmental adaptability of the measured tree, multiple spectral images of the measured tree can be obtained through remote sensing technology. These spectral images usually include green spectrum, red spectrum, red edge spectrum, infrared spectrum and white light spectrum. The green spectrum refers to the spectrum in the green band range; the red spectrum refers to the spectrum in the red band range; the red edge spectrum refers to the spectrum between the red and near-infrared bands; the infrared spectrum refers to the spectrum in the near-infrared band range; and the white light spectrum refers to the spectrum in the visible light band range, which mainly reflects the color and morphology of the plant. Multiple spectra need to be recorded at the same time and save the time point information.

[0060] Step 103, pixel cropping is performed on the multispectral images to remove the areas where the image edge features do not match, and a multispectral image set with completely coinciding features is obtained.

[0061] Specifically, after obtaining the multispectral images, due to the spectral response characteristics of different bands, there may be areas where the image edge features do not match, which will affect the subsequent image processing and analysis, so these areas need to be removed. In order to obtain a multispectral image set with completely coinciding features, pixel cropping can be performed on the images. In one embodiment, a reference image is selected, and then the other images are compared with the reference image to find the mismatched areas between them, and these areas are cropped from the images to obtain a multispectral image set with completely coinciding features.

[0062] Step 104, image normalization calculation is performed, and a normalized image is regenerated according to the normalization calculation result.

[0063] Specifically, when processing images, normalization processing is usually needed to better perform subsequent processing. Image normalization is to scale the pixel values of the image to a specific range, usually [0, 1] or [-1, 1]. At the same time, a normalized image can also be regenerated according to the normalization calculation result, and the normalized pixel values are scaled back to the original pixel value range.

[0064] In the above step S102, the green spectrum image is denoted as Green, the red spectrum image is denoted as Red, the red edge spectrum image is denoted as RedEdge, and the infrared spectrum image is denoted as NIR;

[0065] NDVI = (NIR-Red) / (NIR+Red);

[0066] NDRE = (NIR-RedEdge) / (NIR+RedEdge);

[0067] OSAVI = (NIR-Red) / (NIR+Red+0.16);

[0068] LCI = (NIR-RedEdge) / (NIR+Red);

[0069] GNDVI = (NIR-Green) / (NIR+Green);

[0070] According to the normalized calculation results, the normalized images are regenerated, which are NDVI image, NDRE image, OSAVI image, LCI image and GNDVI image respectively.

[0071] Step 105, the pixel numbers of tree height H, crown width A and diameter at breast height C are obtained, and the actual values of tree height H, crown width A and diameter at breast height C are calculated.

[0072] Step 106, the RGB feature information of tree leaves is obtained, and the specific growth period T of the measured tree is determined.

[0073] Specifically, the position of the maximum normalized value is found in the NDVI image in step S104, which is the position of the tree leaves, the RGB feature information of the white light spectrum of the tree leaves of the measured tree is obtained, and the RGB feature information is matched with the pre-stored RGB feature information of each tree growth period to obtain the specific growth period T of the tree.

[0074] Step 107, the growth data of the measured tree in a year is obtained to obtain the growth model Φ of the measured tree; the generated data includes tree height H, crown width A, diameter at breast height C and life cycle T.

[0075] Specifically, the growth data of a tree in a year is obtained, including tree height, crown width, diameter at breast height and life cycle. Through these data, the growth model Φ of the tree can be obtained by using the method of polynomial fitting. Model Φ can help us better understand the growth law of the tree, and provide reference for the management and protection of the tree. The purpose of polynomial fitting is to find a polynomial function that is closest to the given data points. In practical application, high-order (more than three) polynomial functions can be used to fit the data.

[0076] The present application can quickly obtain multi-spectral images of trees, detect tree growth data, and construct a tree growth model, which can be widely used in smart forestry, ecological environmental protection, forestry carbon sink and other fields.

[0077] As a specific embodiment of a forest annual growth data remote sensing method based on image recognition, the method for obtaining the pixel numbers of tree height H, crown width A and diameter at breast height C includes:

[0078] Seven key points and a center line L are selected; the first key point o1 is the center point of the root of the measured tree; the center line L is a straight line passing through o1 and perpendicular to the ground; the second key point o2 is the top point of the contour of the measured tree; the third key point o3 is the left edge point of the contour of the measured tree farthest from the center line; the fourth key point o4 is the right edge point of the contour of the measured tree farthest from the center line; the fifth key point o5 is a point on the center line at a fixed distance from o1; the sixth point o6 is the left contour point farthest from the center line at the same height as o5; and the seventh point o7 is the right contour point farthest from the center line at the same height as o5.

[0079] wherein:

[0080] The pixel number m of the tree height H is o2-o1;

[0081] The pixel number n of the crown width A is o4-o3;

[0082] The pixel number t of the breast diameter C is o7-o6.

[0083] Specifically, the above-mentioned measurement method does not need to contact the measured tree, avoiding the damage to the tree caused by the traditional measurement method; by selecting seven key points and a center line L, the pixel numbers of the tree height H, the crown width A and the breast diameter C can be accurately measured, so as to realize the accurate monitoring and analysis of the growth state of the tree.

[0084] As a specific embodiment of the image recognition-based forest tree all-year growth data remote measurement method, the pixel numbers of the tree height H, the crown width A and the breast diameter C are obtained and the actual values of the tree height H, the crown width A and the breast diameter C are calculated, comprising:

[0085] Selecting a normalized image with the clearest contour and the highest contrast;

[0086] Randomly selecting a cluster center and classifying all sample points according to the current cluster center;

[0087] Calculating the mean value of the sample points of each class as the cluster center of the next iteration;

[0088] Calculating the difference between the cluster center of the next iteration and the current cluster center;

[0089] If the difference is less than a given iteration threshold, the iteration ends, otherwise the iteration continues; arranging the segmentation targets of the cluster in the order from top to bottom and from left to right to form the contour of the measured tree;

[0090] According to the corresponding relationship between the pixel size and the actual size, the actual values of the tree height H, the crown width A and the breast diameter C can be obtained.

[0091] Specifically, according to the correspondence between the pixel size and the actual size, the actual values of the tree height H, the crown width A, and the breast diameter C can be obtained. For example, an object with a known height Y and width X is placed in the scene, the number of pixels y and x of the object in the height and width directions of the image are measured, and the actual value corresponding to 1 pixel in the height direction is Y / y, and the actual value corresponding to 1 pixel in the width direction is X / x; wherein the tree height H is the value in the height direction, and the crown width A and the breast diameter C are the values in the width direction.

[0092] The actual values of the tree height H, the crown width A, and the breast diameter C can be calculated by the following formulas:

[0093] Tree height H = m x Y / y;

[0094] Crown width A = n x X / x;

[0095] Breast diameter C = t x X / x.

[0096] As a specific embodiment of the image recognition-based forest tree annual growth data remote sensing method, the application further provides a multispectral image recognition device, which comprises a visual camera and a holder.

[0097] The holder comprises a posture reading assembly, a stepping motor, and a digital processing unit; the digital processing unit comprises a posture control module and a spectrum analysis module.

[0098] The visual camera outputs optical images to the digital processing unit, and different spectrum filters 3 in the visual camera output images of different spectrums.

[0099] The posture reading assembly outputs the vertical rotation angle and the horizontal rotation angle of the holder to the digital processing unit.

[0100] The stepping motor comprises a vertical stepping motor 10 and a horizontal stepping motor 16 to drive the holder to rotate vertically and horizontally.

[0101] The posture control module sends instructions to the stepping motor to make the stepping motor rotate in at least one of the vertical angle and the horizontal angle.

[0102] The spectrum analysis module is used to process the spectrum data at different vertical angles and horizontal angles, and splice the full-spectrum image of the measured scene.

[0103] Specifically, the spectral analysis module processes the spectral data of the multispectral vision camera matrix at different vertical and horizontal angles, and splices the panoramic spectral image of the measured scene. By comparing and analyzing different characteristic spectra, specific targets in the measured scene can be extracted, and physical data, biological data and chemical data related to the target can be measured. Through the attitude control of the holder and the driving of the stepper motor, the measured scene can be shot in all directions, so as to obtain high-precision spectral data. The filter 3 inside the vision camera can output images of different spectra, thereby realizing the collection of multispectral data. Through the processing of the spectral analysis module, the spectral data at different vertical and horizontal angles can be spliced into a panoramic spectral image of the measured scene, thereby realizing the remote measurement of the annual growth data of the forest.

[0104] As one of the implementation modes of the image recognition-based forest annual growth data remote measurement method, the vision camera comprises a lens 1 and a camera 2, the camera 2 is fixedly connected with the lens 1, and the filter 3 is installed inside the lens 1.

[0105] As one of the implementation modes of the image recognition-based forest annual growth data remote measurement method, the holder comprises a base 4 for installing the vision camera, one end of the base 4 is provided with an observation port 5, the lens 1 is aligned with the observation port 5, and the camera 2 obtains external images and optical information from the observation port 5 through the lens 1.

[0106] As one of the implementation modes of the image recognition-based forest annual growth data remote measurement method, the holder further comprises: a horizontal base 6 and a vertical shaft 7 rotatably installed at the lower end of the base 6; a rotating platform 8 is fixedly sleeved on the outer circumference of the vertical shaft 7; a vertical shaft encoder 9 is arranged at the upper end of the vertical shaft 7, and the vertical shaft encoder 9 detects and outputs the horizontal rotation angle of the vision camera; the vertical shaft 7 is connected with a vertical stepper motor 10, and the vertical stepper motor 10 drives the vertical shaft 7 to rotate.

[0107] As one of the implementation modes of the image recognition-based forest annual growth data remote measurement method, the rotating platform 8 is fixedly installed with a lower frame 11; the lower frame 11 is in an inverted U shape; a first horizontal shaft 12 and a second horizontal shaft 13 are installed on the two sides of the lower frame 11 and can rotate on the lower frame 11;

[0108] The two ends of the first horizontal shaft 12 and the second horizontal shaft 13 are fixedly connected with the two sides of an inverted upper frame 14, respectively;

[0109] The base 4 is fixedly installed on the upper surface of the upper frame 14;

[0110] One end of the first horizontal shaft 12 is provided with a horizontal shaft encoder 15, and one end of the second horizontal shaft 13 is connected with a horizontal stepper motor 16; the horizontal stepper motor 16 drives the first horizontal shaft 12 and the second horizontal shaft 13 to rotate;

[0111] The horizontal shaft code disc 15 detects and outputs the vertical rotation angle of the vision camera.

[0112] As one of the implementation manners of the image recognition-based forest growth data remote measurement method, the posture control module sends instructions to the stepper motor to rotate the horizontal angle and the vertical angle, so as to drive the horizontal shaft and the vertical shaft 7 to rotate to the required horizontal angle and vertical angle.

[0113] The spectrum analysis module processes the spectrum data of the multi-spectrum vision camera matrix at different vertical angles and horizontal angles to generate a panoramic spectrum image of the measured scene.

[0114] The image stitching algorithm, the frame synchronization algorithm and the multi-spectrum image processing algorithm are pre-stored in the spectrum analysis module.

[0115] The image stitching algorithm is to stitch several images obtained by different vision cameras at different angles into a seamless panoramic image. Since the horizontal angle and the vertical angle of each vision camera can be accurately measured, the stitching is very efficient, and only a small overlap area is required between images.

[0116] The frame synchronization algorithm is to realize the frame synchronization of multiple vision devices in the local area network by using the NTP synchronization protocol. The algorithm estimates the time broadcasted from the client to the server and calculates the waiting time to accurately estimate the time difference between the vision devices, so as to realize the frame synchronization of the multiple vision devices, thereby ensuring that each image participating in the stitching is obtained at the same time.

[0117] The multi-spectrum image processing algorithm is to calculate the panoramic images of different spectrums, including the calculation of the normalized index, the color change and the color inverse transformation, the extraction of specific targets corresponding to the characteristic spectrum in the panoramic image, and the measurement of the physical data, biological data and chemical data related to the target.

[0118] It should be noted that the above examples are only used to illustrate the present application and not to limit the technical solutions described in the present application. Although the present application has been described in detail with reference to the above examples, those skilled in the art should understand that the skilled in the art can still modify or equivalently replace the present application, and all technical solutions and improvements within the spirit and scope of the present application should be covered in the scope of the claims of the present application.

Claims

1. A method for remotely measuring the annual growth data of forest trees based on image recognition, characterized in that, include: The multispectral image recognition device, under the control of the positioning algorithm, is aligned with the tree being measured and records the current coordinates of the tree. Acquire images of the tree under test using multiple spectra; Multiple spectra are recorded and time-point information is stored at the same time; the multispectral images are cropped pixel by pixel to remove mismatched edge features, resulting in a set of multispectral images with completely overlapping features; image normalization is calculated, and a normalized image is regenerated based on the normalization result; the number of pixels for tree height H, crown width A, and diameter at breast height C is obtained, and the actual values ​​of tree height H, crown width A, and diameter at breast height C are calculated; the RGB feature information of the leaves is obtained to determine the specific growth cycle T of the measured tree; Furthermore, the growth data of the tested tree throughout the year is fitted to obtain the growth model Φ of the tested tree; the generated data includes: tree height H, crown width A, diameter at breast height C, and lifespan T; the method for obtaining the number of pixels of tree height H, crown width A, and diameter at breast height C includes: selecting seven key points and a center line L; the first key point o1 is the center point of the root of the tested tree; the center line L is a straight line passing through o1 and perpendicular to the ground; the second key point o2 is the vertex of the outline of the tested tree; the third key point o3 is the left edge point of the outline of the tested tree that is farthest from the center line; the fourth key point o4 is the right edge point of the outline of the tested tree that is farthest from the center line; the fifth key point o5 is a point on the center line at a fixed distance from o1; the sixth point o6 is the left outline point farthest from the center line at the same height as o5; the seventh point o7 is the right outline point farthest from the center line at the same height as o5; where: the number of pixels of tree height H m = o2 - o1; the number of pixels of crown width A n = o4 - o3; The number of pixels for diameter at breast height (DBH) C is t = 07 - 06; obtaining the number of pixels for tree height (H), crown width (A), and DBH C, and calculating the actual values ​​of tree height (H), crown width (A), and DBH C, includes: selecting the normalized image with the clearest outline and highest contrast; randomly selecting cluster centers and classifying all sample points according to the current cluster center; calculating the mean of the sample points in each class as the cluster center for the next iteration; calculating the difference between the cluster center of the next iteration and the current cluster center; if the difference is less than a given iteration threshold, the iteration ends; otherwise, the iteration continues; arranging the clustering targets in order from top to bottom and from left to right to form the outline of the measured tree; based on the correspondence between pixel size and actual size, the tree height (H), crown width (A), and DBH C, and their actual values, can be obtained.

2. A multispectral image recognition device for implementing the remote sensing method for annual tree growth data as described in claim 1, characterized in that, include: A visual camera and a gimbal; wherein the gimbal includes an attitude reading component, a stepper motor, and a digital processing unit; the digital processing unit includes an attitude control module and a spectral analysis module; the visual camera outputs optical images to the digital processing unit, and the visual camera has filters (3) of different spectra inside to output images of different spectra; the attitude reading component outputs the vertical rotation angle and horizontal rotation angle of the gimbal to the digital processing unit; the stepper motor includes a vertical stepper motor (10) and a horizontal stepper motor (16) to drive the gimbal to rotate vertically and horizontally; the attitude control module sends commands to the stepper motor to cause the stepper motor to rotate at least one of the vertical and horizontal angles; the spectral analysis module is used to process spectral data at different vertical and horizontal angles and stitch them together to form a panoramic spectral image of the scene under test.

3. The multispectral image recognition device according to claim 2, characterized in that, The vision camera includes a lens (1) and a camera (2), the camera (2) is fixedly connected to the lens (1), and the filter (3) is installed inside the lens (1).

4. The multispectral image recognition device according to claim 3, characterized in that, The gimbal includes a base (4) for mounting the vision camera. One end of the base (4) is provided with an observation port (5). The lens (1) is aligned with the observation port (5). The camera (2) obtains external images and optical information from the observation port (5) through the lens (1).

5. The multispectral image recognition device according to claim 4, characterized in that, The gimbal also includes: a horizontally placed base (6) and a vertical shaft (7) rotatably mounted on the base (6) at its lower end; a rotating platform (8) is fixedly sleeved on the outer circumference of the vertical shaft (7); a vertical shaft encoder (9) is provided at the upper end of the vertical shaft (7), and the vertical shaft encoder (9) detects and outputs the horizontal rotation angle of the vision camera; the vertical shaft (7) is connected to the vertical stepper motor (10), and the vertical stepper motor (10) drives the vertical shaft (7) to rotate.

6. The multispectral image recognition device according to claim 5, characterized in that, A lower frame (11) is fixedly installed on the rotating platform (8); the lower frame (11) is inverted U-shape; a first horizontal shaft (12) and a second horizontal shaft (13) are installed on both sides of the lower frame (11) and can rotate on the lower frame (11); the two ends of the first horizontal shaft (12) and the second horizontal shaft (13) are respectively fixedly connected to the two sides of the inverted upper frame (14); the base (4) is fixedly installed on the upper surface of the upper frame (14); a horizontal axis encoder (15) is provided at one end of the first horizontal shaft (12), and one end of the second horizontal shaft (13) is connected to a horizontal stepper motor (16); the horizontal stepper motor (16) drives the first horizontal shaft (12) and the second horizontal shaft (13) to rotate; the horizontal axis encoder (15) detects and outputs the vertical rotation angle of the vision camera.

7. The multispectral image recognition device according to claim 6, characterized in that: The attitude control module sends commands to the stepper motor to rotate it at appropriate horizontal and vertical angles, thereby driving the corresponding horizontal and vertical axes (7) to rotate to the required horizontal and vertical angles; the spectral analysis module processes the spectral data of the vision camera at different vertical and horizontal angles to generate a panoramic spectral image of the scene under test; the spectral analysis module has pre-stored image stitching algorithms, frame synchronization algorithms and multispectral image processing algorithms.

Images