Method for measuring and system for predicting thick growth of trees

Abstract

To provide a method for measuring and a system for predicting thick growth of trees with which it is possible to easily provide higher measurement accuracy in measuring the height of a tree from an image obtained by imaging the tree with a camera.SOLUTION: A method for measuring thick growth of trees is to measure the tree height of a tree to be measured from an image obtained by imaging the tree with a camera, with a panel installed near the tree as the background. The method includes: an image forming step of cutting out, from the image, an image to be analyzed in an area corresponding to the panel; a tree image extraction step of extracting, from the image to be analyzed, a tree image area corresponding to the tree through image recognition using segmentation; and a mask creation step of extracting a contour image area including at least the entirety of the trees from a difference in the contrast of the image to be analyzed to create a mask, and the tree image extraction step includes providing the mask to the image to be analyzed to perform image recognition in the contour image area. The system has a program including the steps.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present invention relates to a tree growth measurement method and prediction system using image processing, and more particularly to a tree growth measurement method and prediction system that measures the height of a tree from an image of the tree taken by a camera. [Background technology]

[0002] Solar panels are installed in various locations, and it is necessary to manage the weeds, plants, or trees (hereinafter, collectively referred to as "trees") around the solar panels so as not to reduce power generation efficiency or spoil the scenery. Generally, trees are manually cut down as needed so that their height does not exceed the height of the installed solar panels. However, because the rate at which trees grow is highly dependent on the climate, it is also necessary to check the state of tree growth at the site where the solar panels are installed. Therefore, cameras are installed mainly in remote locations and mountainous areas with poor accessibility, and remote inspections are performed. Furthermore, to reduce labor, a system has been proposed that processes camera images to automatically determine the state of growth.

[0003] For example, Patent Document 1 discloses a tree management system that uses a stereo camera to capture images of target trees and estimate their position and shape. It claims that the image information from the stereo camera can be used to extract the target tree region from a background image based on image recognition using semantic segmentation based on differences in shooting distance. The target tree region generally consists of a relatively thin lower branch (trunk) and an upper, roughly triangular crown, which can be recognized through image recognition. The system then estimates the distance between the left and right measurement points from the geometric relationship between the angle between the camera and the lines pointing to the left and right measurement points and the shooting distance, and also estimates the tree height from the corresponding number of pixels. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2023-105664 Summary of the Invention [Problem to be solved by the invention]

[0005] When attempting to directly recognize target tree regions using image segmentation, such as by using deep learning image recognition, it is difficult to achieve image recognition that provides practical measurements of tree height due to the wide variety of tree-like shapes. In particular, it was found that dependence on training data has a significant impact on recognition accuracy. Therefore, a simple method that can achieve higher measurement accuracy was sought.

[0006] The present invention has been made in consideration of the above-mentioned circumstances, and its purpose is to provide a tree growth measurement method and prediction system that can easily and with higher measurement accuracy when measuring the height of a tree from an image of the tree taken with a camera. [Means for solving the problem]

[0007] The method according to the present invention is a tree growth measurement method that measures the height of a tree to be measured from an image captured by a camera with the tree against a background of a panel installed near the tree, and includes an image formation step that cuts out an analysis target image of a region corresponding to the panel from the image, and a tree image extraction step that extracts a tree image region corresponding to the tree from the analysis target image by image recognition using segmentation, and includes a mask generation step that extracts a contour image region that encompasses at least the entire tree from the contrast difference of the analysis target image and generates a mask, and the tree image extraction step is characterized in that the mask is applied to the analysis target image and the image recognition is performed in the contour image region.

[0008] According to this feature, by using a mask obtained from optical differences, it becomes possible to easily obtain tree shapes through image recognition with higher measurement accuracy, even for objects such as trees, which have a wide variety of shapes.

[0009] In the above-described invention, the panel may be colored to provide the background with a combination of multiple colors. Alternatively, the panel may be painted with horizontally extending monochromatic stripes, each of which is painted with one of the multiple colors. Furthermore, the panel may be painted with white stripes at regular intervals on a blue background. According to this feature, masks can be obtained optically with high accuracy even for objects such as trees, which have a wide variety of shapes, resulting in simple and high measurement accuracy.

[0010] The above-described invention may further include a correction step of correcting focus of the image to be analyzed prior to the segmentation process. The correction step may be edge correction or histogram equalization. According to this feature, a mask can be accurately obtained by segmentation even for an object with a wide variety of shapes, such as a tree, resulting in a simple and highly accurate measurement.

[0011] In the above-described invention, the image recognition may be performed by deep learning. According to this feature, the influence of learning data obtained by deep learning can be suppressed, and as a result, it is possible to easily achieve higher measurement accuracy.

[0012] In the above-described invention, the panel may have a known panel height from the ground, and the tree height may be measured from the ratio of the maximum height position of the tree image area to the image to be analyzed. With this feature, the tree height can be calculated easily and with higher measurement accuracy.

[0013] In addition, the system according to the present invention is a tree growth measurement system that measures the height of a tree to be measured from an image captured by a camera with the tree against a background of a panel installed near the tree, and has a program including an image formation step that cuts out an analysis target image of an area corresponding to the panel from the image, and a tree image extraction step that extracts a tree image area corresponding to the tree from the analysis target image by image recognition using segmentation, and includes a mask generation step that extracts a contour image area that encompasses at least the entire tree from the contrast difference of the analysis target image and generates a mask, and the tree image extraction step is characterized in that the mask is applied to the analysis target image and the image recognition is performed in the contour image area.

[0014] According to this feature, by using a mask obtained from optical differences, it becomes possible to easily obtain tree shapes through image recognition with higher measurement accuracy, even for objects such as trees, which have a wide variety of shapes.

[0015] In the above-described invention, the panel may be colored to provide the background with a combination of multiple colors. Alternatively, the panel may be painted with horizontally extending monochromatic stripes, each of which is painted with one of the multiple colors. Furthermore, the panel may be painted with white stripes at regular intervals on a blue background. According to this feature, masks can be obtained optically with high accuracy even for objects such as trees, which have a wide variety of shapes, resulting in simple and high measurement accuracy.

[0016] The above-described invention may further include a correction step of correcting focus of the image to be analyzed prior to the segmentation process. The correction step may be edge correction or histogram equalization. According to this feature, a mask can be accurately obtained by segmentation even for an object with a wide variety of shapes, such as a tree, resulting in a simple and highly accurate measurement.

[0017] In the above-described invention, the image recognition may be performed by deep learning. According to this feature, the influence of learning data obtained by deep learning can be suppressed, and as a result, it is possible to easily achieve higher measurement accuracy.

[0018] In the above-described invention, the panel may have a known panel height from the ground, and the tree height may be measured from the ratio of the maximum height position of the tree image area to the image to be analyzed. With this feature, the tree height can be calculated easily and with higher measurement accuracy. [Brief explanation of the drawings]

[0019] [Figure 1] 1 is a block diagram of an embodiment of a tree growth measurement system according to the present invention. [Figure 2] 1 is an example of an image used in the tree growth measurement method according to the present invention. [Figure 3] FIG. 10 is a flow diagram of advance preparation for the measurement system. [Figure 4] FIG. 1 is a flow chart showing a part of an embodiment of a method for measuring tree growth. [Figure 5] FIG. 1 is a flow chart showing a part of an embodiment of a method for measuring tree growth. [Figure 6] (a) Original image, (b) image to be analyzed, (c) inverted image, and (d) mask used for overgrowth measurement. [Figure 7](a) Brightness-edge adjusted image, (b) edge adjusted image, (c) contrast adjusted image, and (d)-(f) tree-defined images with different contrast adjustment paths. [Figure 8] Other images used for overgrowth measurement are (a) original image, (b) image to be analyzed, (c) inverted image, and (d) mask. [Figure 9] Other images with different contrast adjustment paths are (a) brightness-edge adjusted image, (b) edge adjusted image, (c) contrast adjusted image, and (d) to (f) tree-defined images. [Figure 10] Further images used for overgrowth measurement are (a) original image, (b) image to be analyzed, (c) inverted image, and (d) mask. [Figure 11] Further images with different contrast adjustment paths are (a) brightness-edge adjusted image, edge adjusted image, (c) contrast adjusted image, and (d) to (f) tree-defined images. DETAILED DESCRIPTION OF THE INVENTION

[0020] A tree growth measurement method and a tree growth prediction system used therefor as one embodiment of the present invention will be described with reference to FIGS. 1 to 11. FIG.

[0021] As shown in FIG. 1, measurement system 10 includes server 1 equipped with database 2 for storing images and other information used in overgrowth measurement. Server 1 is connectable to a terminal 4 used by a user of measurement system 10 via a network 3 such as the Internet, and can execute operations based on a predetermined program in response to operations on terminal 4 and output measurement results to terminal 4. Server 1 is also connected via network 3 to a camera 5 installed so as to capture a tree 6 (weeds, plants, or trees) to be measured within its field of view. A panel 7 is installed within the field of view of camera 5, forming a background for tree 6. This allows camera 5 to capture an image of panel 7 along with tree 6 to be measured, thereby obtaining an image including panel 7 as a background for tree 6. The obtained image is collected by server 1 via network 3 and stored in database 2.

[0022] Referring also to FIG. 2, the panel 7 is installed, for example, in front of the solar panel 8. In the growth measurement, the main objective is to keep the height of the tree 6 lower than the height of the solar panel 8 so that the tree 6 does not cast a shadow on the solar panel 8. Therefore, for example, it is conceivable to measure the height of the tree 6 near the lower edge of the inclined light-receiving surface of the solar panel 8, or to select a location around the solar panel 8 where the tree 6 is most likely to grow and measure its height. Based on this, the position of the tree 6 to be measured is determined, the panel 7 is installed near that location, and the camera 5 is positioned so that the panel 7 is included as a background to the tree 6 in the obtained image. In other words, the camera 5 is installed facing the panel 7 on an extension line connecting the panel 7 and the tree 6 to be measured. The same figure shows an example in which the panel 7 is installed in front of the lower edge of the inclined surface of the solar panel 8 and is lower in height than that edge.

[0023] To facilitate measurement of the height of the tree 6, the panel 7 is designed to stand out against the surrounding scenery and highlight the tree 6 against the background of the panel 7. For example, a color that contrasts relatively highly with the green color of the leaves of the tree 6 is preferable. Therefore, for example, the panel 7 may be colored using a combination of multiple colors. This makes it easier to make the background of the tree 6 stand out against the background and highlight the tree 6. The panel 7 also has a known panel height and is used as a ruler on the image when measuring the height of the tree 6. For example, the height of the tree 6 can be measured in relation to the panel height on the obtained image. Alternatively, the panel 7 may have horizontally extending monochromatic stripes, each of which may be colored one of multiple colors. Here, a vertically elongated rectangular panel 7 with regularly spaced white stripes on a blue background was used. Horizontal stripes can also be used to measure the height of the tree 6.

[0024] As shown in FIG. 3, advance preparation (S10) is performed on the server 1 of the measurement system 10. First, a basic setting file is loaded (S11). The basic setting file contains setting values ​​related to processing common to all images, such as the size of the image after cropping and projective transformation of the area corresponding to the panel 7 (described later), setting values ​​for brightness correction processing, designation of the color of the panel 7, mask generation, edge correction, CLAHE processing, and other image processing settings. Next, a power plant setting file is loaded (S12). The power plant setting file contains setting values ​​for the height of the panel 7, the path indicating the storage location of the template image, the reference position within the template image, and the cropping position, as well as thresholds and tolerances for template matching.

[0025] Next, a template is loaded (S13). The template defines the shape and color scheme of the panel 7 and is used to extract the area of ​​the panel 7 from the obtained image by template matching, as described below. Therefore, the shape and color scheme of the panel 7 are designed to be unique to nature, with the rectangular blue and white stripes described above. Furthermore, a segmentation model is loaded (S14). In image recognition using segmentation, as described below, a tree image region corresponding to the tree 6 is extracted. Therefore, this segmentation model is a model constructed by deep learning so as to enable extraction of a region corresponding to the tree 6. Here, because the tree 6 is extracted by segmentation, a model that can recognize that it is some kind of plant regardless of color is preferable; for example, a model based on a panoptic segmentation technique is preferable.

[0026] The server 1 then acquires images taken at a predetermined time by a predetermined camera 5 via the network 3 and stores them in the database 2. At this time, the file names of the images contain codes that allow for obtaining various information such as the location and time the image was taken. Meanwhile, based on instructions from the terminal 4, the server 1 performs overgrowth measurement on the specified image from among the stored images. The overgrowth measurement involves the following process.

[0027] 4 and 5, along with FIGS. 6 and 7, in the growth measurement (S20), first, an image file captured by camera 5 is selected based on an instruction from terminal 4, and various information is extracted from the file name (S21). For example, a unique identification number or other code may be included in the file name, and a table of various information corresponding to this code may be created. The image of the selected file is then loaded (S22) to create an original image 11 (see FIG. 6(a)) for use in image analysis. Template matching is performed on original image 11 (S23), and an area of ​​the image corresponding to panel 7 is extracted and cut out. Projective transformation is then performed to correct the shape of panel 7 for trapezoidal correction so that the shape of panel 7 becomes a predetermined rectangle (S24). In template matching, a template is used with the top edge of panel 7 as the reference position, preventing the reference position of the template from being hidden by trees 6. In this way, the area corresponding to panel 7 is extracted and cut out as an image to be analyzed 12 (see FIG. 6(b)) (image formation steps: S23 to S24).

[0028] The cut-out analysis target image 12 is used to extract a tree image region corresponding to the tree 6 by image recognition using segmentation (S30; see Figure 5), which will be described later, and is also used to generate a mask (mask generation steps S25 to S27).

[0029] To generate a mask, first, the brightness of the analysis target image 12 is corrected (S25). As shown in FIG. 6(a), the brightness correction involves correcting the brightness of the entire analysis target image 12 so that the average brightness of the analysis target image 12 matches the set value for the brightness correction process described above, thereby reducing variation between images. Then, portions that do not correspond to the panel 7 (portions that are not blue-white) are extracted by inverting the blue-white color (S26). Specifically, portions of the analysis target image 12 that fall within the color value ranges set for blue and white are first extracted, and the extraction results are inverted to extract portions that do not correspond to blue or white. The inverted image 13 (see FIG. 6(c)) corresponding to this extraction result is a binary image in which extracted portions that do not correspond to blue or white are displayed in white, and extracted portions that do not correspond to blue or white are displayed in black.

[0030] In the image to be analyzed 12, there is a blue-white portion corresponding to panel 7 and a portion corresponding to tree 6, and in an inverted image 13 in which the portion corresponding to tree 6 has been extracted, the portion corresponding to tree 6 is displayed in white. Furthermore, a contour is defined between the extracted region corresponding to tree 6 and its surroundings so as to exclude regions unlikely to correspond to tree 6, and a mask 14 (see FIG. 6(d)) is created (S27). At this time, there is a portion corresponding to tree 6 within the range of the defined contour, but the contour is set wider so as to include the portion corresponding to the entire tree 6. Specifically, a contour image region 15 is defined as mask 14 so as to exclude only the region that is clearly blue-white and corresponds to panel 7, that is, the region that is clearly blue or white and does not correspond to tree 6.

[0031] In the inverted image 13 obtained by blue-white inversion extraction (S26), the height of the portion (white portion) extracted as corresponding to the tree 6 is measured and set as the first tree height (S28). As described above, the height of the tree 6 is measured as a ratio to the panel height. Up to this point, image processing based on optical differences such as color and brightness has been performed on the image 12 to be analyzed, and a mask 14 has been created and the first tree height has been measured.

[0032] Next, in the image 12 to be analyzed, the height of the tree 6 is measured by image recognition using segmentation (S30). Here, the image 12 to be analyzed is processed by three routes with different pre-processing steps for performing segmentation.

[0033] First, in the first path A, the analysis target image 12 undergoes correction steps of brightness correction (S31) and edge correction (S32) to produce a brightness / edge-adjusted image 16a (see Figure 7(a)), which is then segmented (S33). Brightness correction is a process of adjusting the average brightness, as described above. Edge correction is a focus correction process that corrects the color (RGB values) to emphasize the edge portion to correct so-called out-of-focus images. In segmentation, all pixels in the brightness / edge-adjusted image 16a of the analysis target image 12 are labeled so that the area corresponding to the tree 6 can be extracted through image recognition in subsequent processing. As described above, segmentation uses a method based on a model capable of extracting trees through deep learning.

[0034] In the second route B, brightness correction is not performed as preprocessing, but rather a correction step using edge correction (S34) is performed to obtain an edge-adjusted image 16b (see FIG. 7(b)), after which segmentation is performed (S35).

[0035] In the third path C, a correction step using histogram equalization correction (S36) is performed to locally equalize the contrast histogram using CLAHE (Contrast Limited Adaptive Histogram Equalization), and then a contrast-adjusted image 16c (see Figure 7(c)) is obtained, followed by segmentation (S37).

[0036] In this way, by adjusting contrast and color using different correction steps for each of the three paths before performing segmentation, it is possible to complement each other's weaknesses in the processing of each path that may arise due to factors such as the amount of light in the image being analyzed.

[0037] The analysis target image 12 (brightness / edge-adjusted image 16a, edge-adjusted image 16b, contrast-adjusted image 16c) segmented along the three paths is combined with the mask 14, and a tree image region 18 is defined by image recognition (tree image extraction steps: S41a, S41b, S41c). Here, the range for image recognition is limited to within the contour image region 15 of the mask 14, and the portion corresponding to the tree 6 is extracted by image recognition to obtain tree demarcated images 17a-17c as the tree image region 18. In other words, image recognition is performed to determine the portion corresponding to the tree 6 from the labels attached to the pixels, and the pixels corresponding to the tree 6 are extracted to create the tree demarcated images 17a-17c. Then, in each of the tree demarcated images 17a-17c, the maximum height of the tree image region 18 is measured as the height of the tree 6 (S42a-S42c), and the maximum value of the three measured heights is calculated (S43) and designated as the second tree height. The reason for using the maximum value as the second tree height is to obtain a result on the safe side (higher) in measuring overgrowth. Note that tree delineation images 17a to 17c are originally binary images in which the tree image area is white and the other areas are black. Here, to make it easier to compare the analysis target image 12 with the results obtained by segmentation, the transparency of the original tree delineation image was set to 50%, and it was then overlaid on each of brightness / edge-adjusted image 16a, edge-adjusted image 16b, and contrast-adjusted image 16c to display them as tree delineation images 17a to 17c.

[0038] The first tree height measured by optical processing (see S28) is then compared with the second tree height measured by image recognition (S44). If the difference between the first tree height and the second tree height is considered small (for example, when twice the first tree height is greater than the second tree height), the higher value is adopted. If the difference is large (for example, when twice the first tree height is equal to or less than the second tree height), the first tree height is adopted, and the result is output as the tree height (S45). The reason for adopting the higher value when the difference is small is to obtain a safe result in overgrowth measurement, as described above. On the other hand, the reason for adopting the first tree height when the difference is large is to anticipate a case in which a misrecognition occurs in image recognition using segmentation, resulting in the second tree height being determined to be an abnormal value. In other words, if either or both of the first tree height and the second tree height contain abnormal values, the first tree height is adopted, which allows for a relatively stable height measurement. The value to be adopted can be determined as appropriate.

[0039] In this way, tree height can be measured. Note that in overgrowth measurement, it is necessary to recognize "trees" that encompass a wide variety of plants, rather than recognizing plants of a specific shape. In other words, there is a large variation in the shape and color of trees. For this reason, in this embodiment, panel 7 is used as a background that is clearly not a tree, and the background is removed, and ambiguous areas are determined using deep learning.

[0040] In the above example, original image 11 was captured in a backlit position, as can be seen from the sun in the upper right corner. This made it relatively difficult to image-recognize the portion corresponding to tree 6. For example, the actual measured height of tree 6 was 750 mm, but when contrast was adjusted using CLAHE in route C, the measured tree height was 4 mm. In other words, measuring the height of tree 6 through image recognition was relatively difficult. However, by adjusting the brightness and color in routes A and B, tree heights of 730 mm and 728 mm, respectively, were measured. By complementing this, a tree height close to the actual measured value of 730 mm was finally measured. Multiple routes using different image adjustment methods complemented each other, and a mask derived from optical differences was used to prevent false recognition of tree heights that were too high. In other words, even for objects such as trees 6, which have a wide variety of shapes, tree heights can be easily measured with high accuracy through image recognition.

[0041] We will also explain examples of measuring tree heights for other original images.

[0042] As shown in Figure 8(a), original image 11-1 was taken at dusk and has a slightly dark color tone. Similar to the above, template matching (S23) was used to extract and cut out the area corresponding to panel 7 from original image 11-1, and projective transformation was performed (S24) to obtain analysis target image 12-1 (see Figure 8(b)). Furthermore, brightness of analysis target image 12-1 was corrected (S25) to obtain inverted image 13-1 (see Figure 8(c)) through blue-white inversion extraction processing. A contour was defined to exclude areas unlikely to correspond to tree 6, creating mask 14-1 (see Figure 8(d)). The actual height of tree 6 was measured to be 50 mm, and the first tree height measured from inverted image 13-1 was 28 mm.

[0043] Next, as shown in FIG. 9, brightness, color, and contrast were adjusted from the analysis target image 12a via paths A, B, and C to obtain brightness and edge-adjusted image 16-1a, edge-adjusted image 16-1b, and contrast-adjusted image 16-1c, respectively (see FIGS. 9(a), (b), and (c)). Segmentation was then performed (S31-S37). Then, mask 14-1 was combined and tree images were defined using image recognition to obtain tree-defined images 17-1a, 17-1b, and 17-1c (S41a-c). The heights of the trees were measured (S42a-c), and were found to be 56 mm, 58 mm, and 58 mm, respectively. Therefore, the second tree height was determined to be 58 mm (S43). A comparison was made with the first tree height (S44). Since twice the first tree height was smaller than the second tree height, the first tree height was adopted, and the height of tree 6 was output as 28 mm (S45).

[0044] In this way, a tree height of 28 mm was obtained, compared to the actual measured value of 50 mm for tree 6. As mentioned above, this is due to the rule that the first tree height is adopted when there is a large difference between the first tree height and the second tree height, and the second tree height obtained by image recognition is close to the actual measured value. Note that although noise (white areas) occurs at high positions in each of tree definition images 17-1a, 17-1b, and 17-1c, by using mask 14-1, this noise was not mistaken for tree 6, and tree heights could be obtained with high measurement accuracy.

[0045] As shown in Figure 10, tree height was measured using another original image 11-2. Original image 11-2 was also captured at dusk, resulting in a slightly darker color tone. Furthermore, panel 7 was positioned to the left of the field of view. For original image 11-2, template matching (S23) was used to extract and cut out the area corresponding to panel 7, as described above, and projective transformation was performed (S24) to obtain analysis target image 12-2 (see Figure 10(b)). Furthermore, brightness of analysis target image 12-2 was corrected (S25) to obtain inverted image 13-2 (see Figure 10(c)) through blue-white inversion extraction processing. A contour was defined to exclude areas unlikely to correspond to tree 6, creating mask 14-2 (see Figure 10(d)). The actual height of tree 6 was 565 mm, and the first tree height measured from inverted image 13-2 was 530 mm.

[0046] Next, as shown in FIG. 11, the brightness, color, and contrast of the analysis target image 12-2 were adjusted along paths A, B, and C to obtain brightness and edge-adjusted image 16-2a, edge-adjusted image 16-2b, and contrast-adjusted image 16-2c, respectively (see FIGS. 11(a), (b), and (c)). Segmentation was then performed (S31 to S37). Then, the mask 14-2 was combined, and tree images were defined using image recognition to obtain tree-defined images 17-2a, 17-2b, and 17-2c (S41a to S41c). The heights of the respective trees were measured (S42a to S42c), and were found to be 4 mm, 468 mm, and 228 mm, respectively. Therefore, the second tree height was determined to be 468 mm (S43). The heights were compared with the first tree height (S44), and the height of tree 6 was output as 530 mm (S45).

[0047] In this way, we were able to obtain a tree height of 530 mm in original image 11-2, which is close to the actual measured value of 565 mm for the height of tree 6. Although there was noise in this case as well, by using a mask, we were able to obtain a value that was relatively close to the actual measured value for the second tree height using image recognition.

[0048] While typical embodiments of the present invention and modifications based thereon have been described above, the present invention is not necessarily limited to these, and a person skilled in the art will be able to find various alternative embodiments and modifications without departing from the spirit of the present invention or the scope of the appended claims. [Explanation of symbols]

[0049] 1 server 2 Database 5. Camera 6. Trees 7 Panels

Claims

1. A method for measuring the growth of trees, which measures the height of a tree from an image of the tree taken by a camera with a panel installed near the tree as a background, comprising: an image forming step of cutting out an analysis target image of a region corresponding to the panel from the image; a tree image extraction step of extracting a tree image region corresponding to the tree from the analysis target image by image recognition using segmentation, A method for measuring tree growth, comprising a mask generation step of extracting a contour image area encompassing at least the entire tree from the contrast difference of the image to be analyzed and generating a mask, wherein the tree image extraction step applies the mask to the image to be analyzed and performs image recognition in the contour image area.

2. 2. The method for measuring tree growth according to claim 1, wherein the panel is colored to provide the background with a combination of multiple colors.

3. 3. A method for measuring tree growth according to claim 2, wherein the panel is painted with horizontally extending monochromatic stripes, each of which is given one of the plurality of colors.

4. 4. A method for measuring tree growth according to claim 3, wherein the panel has white stripes drawn at regular intervals on a blue background.

5. 2. The method for measuring tree growth according to claim 1, further comprising a correction step of correcting a focus error of the image to be analyzed prior to the segmentation process.

6. 6. The method for measuring tree growth according to claim 5, wherein the correction step is edge correction or histogram equalization correction.

7. The method for measuring tree growth according to any one of claims 1 to 6, characterized in that the image recognition is performed using deep learning.

8. 8. A method for measuring tree growth according to claim 7, wherein the panel has a known panel height from the ground, and the tree height is measured from the ratio of the maximum height position of the tree image area to the image to be analyzed.

9. A tree growth measurement system that measures the height of a tree to be measured from an image of the tree captured by a camera with a panel installed near the tree as a background, an image forming step of cutting out an analysis target image of a region corresponding to the panel from the image; a tree image extraction step of extracting a tree image region corresponding to the tree from the analysis target image by image recognition using segmentation, A tree growth measurement system comprising a mask generation step of extracting a contour image area encompassing at least the entire tree from the contrast difference of the image to be analyzed and generating a mask, wherein the tree image extraction step applies the mask to the image to be analyzed and performs image recognition in the contour image area.

10. 10. The tree growth measurement system according to claim 9, wherein the panel is colored to provide the background with a combination of multiple colors.

11. 11. The tree growth measurement system according to claim 10, wherein the panel is painted with horizontally extending monochromatic stripes, each of the stripes being given one of the plurality of colors.

12. 12. The tree growth measurement system according to claim 11, wherein the panel has white stripes drawn at regular intervals on a blue background.

13. 10. The tree growth measurement system according to claim 9, further comprising a correction step of correcting a focus error of the image to be analyzed prior to the segmentation process.

14. 14. The tree growth measurement system according to claim 13, wherein the correction step is edge correction or histogram equalization correction.

15. The tree growth measurement system according to any one of claims 9 to 14, wherein the image recognition is performed using deep learning.

16. The tree growth measurement system of claim 15, wherein the panel has a known panel height from the ground, and the tree height is measured from the ratio of the maximum height position of the tree image area to the image to be analyzed.

Images