Tree crown delineation method and device

Abstract

The invention relates to the field of forestry remote sensing technology, and provides a tree crown delineation method and device, the method comprising: determining a multi-temporal remote sensing image of a target forest area, the multi-temporal remote sensing image being an image determined based on a plurality of remote sensing images of the target forest area in different seasons; performing tree top area detection on a plurality of filtered images of the multi-temporal remote sensing image respectively, and determining a plurality of first tree top area sets corresponding to the plurality of filtered images respectively; the plurality of filtered images respectively use filters of different sizes; based on multi-temporal spectral information of the multi-temporal remote sensing image, screening and merging a plurality of first tree top areas in a plurality of first tree top area sets, and determining a plurality of second tree top areas; and depicting a tree crown based on the plurality of second tree top areas and a seasonal variation edge image of the multi-temporal remote sensing image, and taking into account the multi-scale characteristics of the tree crown and the multi-temporal spectral factors on the basis of segmenting the multi-temporal image, thereby improving the accuracy of single tree crown delineation.

Description

Technical Field

[0001] The invention relates to the field of forestry remote sensing technology, and in particular to a tree crown delineation method and device. Background Art

[0002] Researching high-precision tree crown delineation algorithms will help to more accurately grasp the dynamic changes of forest crowns, deepen the understanding of forest community succession changes, and provide a theoretical basis for forest management and strengthen the application effectiveness in fine forestry and environmental monitoring. Usually, crown information extraction consists of two steps. The first is single tree detection to obtain the position of the tree top or tree; then single tree crown delineation to automatically obtain the crown boundary. The main methods for single tree detection are local maximum method, multi-scale analysis method and template matching method. The main methods for single tree crown delineation are valley tracking method, regional growth, watershed segmentation and local ray method.

[0003] However, current single-tree detection methods may merge multiple small tree crowns that need to be treated as independent into a large tree crown, resulting in an inability to effectively distinguish different tree crowns more accurately. Summary of the invention

[0004] The present invention provides a tree crown delineation method, device, computer-readable storage medium and electronic device, which can consider the multi-scale characteristics and multi-temporal spectral factors of the tree crown on the basis of multi-temporal images, and perform tree crown screening and merging based on seasonal changes to more accurately retain independent small tree crowns, thereby improving the accuracy of single tree crown delineation and providing technical support for the fine identification and classification of tree species.

[0005] In a first aspect, the present invention provides a tree crown delineation method, comprising:

[0006] Determine a multi-temporal remote sensing image of a target forest area; wherein the multi-temporal remote sensing image is an image determined based on multiple remote sensing images of the target forest area in different seasons;

[0007] Performing tree top region detection on a plurality of filtered images of the multi-temporal remote sensing image, respectively, to determine a plurality of first tree top region sets corresponding to the plurality of filtered images, respectively; wherein the sizes of filters used by the plurality of filtered images are different;

[0008] Based on the multi-temporal spectral information of the multi-temporal remote sensing image, screening and merging a plurality of first tree top regions in the plurality of first tree top region sets to determine a plurality of second tree top regions;

[0009] The tree crown is delineated based on the multiple second tree top areas and the seasonal variation edge image of the multi-temporal remote sensing image to determine the tree crown delineation image.

[0010] In a second aspect, the present invention provides a tree crown delineation device, comprising:

[0011] An image determination unit, used to determine a multi-temporal remote sensing image of a target forest area; wherein the multi-temporal remote sensing image is an image determined based on a plurality of remote sensing images of the target forest area in different seasons;

[0012] A tree top detection unit is used to perform tree top area detection on a plurality of filtered images of the multi-temporal remote sensing image, and determine a plurality of first tree top area sets corresponding to the plurality of filtered images, respectively; wherein the sizes of filters used by the plurality of filtered images are different;

[0013] A screening unit, configured to screen and merge a plurality of first tree top regions in the plurality of first tree top region sets based on the multi-temporal spectral information of the multi-temporal remote sensing image, to determine a plurality of second tree top regions;

[0014] A tree crown depiction unit is used to perform tree crown depiction based on the multiple second tree top areas and the seasonal phase change edge image of the multi-temporal remote sensing image to determine a tree crown delineation image.

[0015] In a third aspect, the present invention provides a computer-readable storage medium, comprising execution instructions. When a processor of an electronic device executes the execution instructions, the processor executes any method described in the first aspect.

[0016] In a fourth aspect, the present invention provides an electronic device, comprising a processor and a memory storing execution instructions, wherein when the processor executes the execution instructions stored in the memory, the processor executes any method described in the first aspect.

[0017] The present invention provides a method, device, computer-readable storage medium and electronic device for delineating a tree crown. The method determines a multi-temporal remote sensing image of a target forest area; wherein the multi-temporal remote sensing image is an image determined by multiple remote sensing images of the target forest area in different seasons; then, tree top area detection is performed on multiple filtered images of the multi-temporal remote sensing image to determine multiple first tree top area sets corresponding to the multiple filtered images; wherein the sizes of filters used by the multiple filtered images are different; then, based on the multi-temporal spectral information of the multi-temporal remote sensing image, multiple first tree top areas in the multiple first tree top area sets are screened and merged to determine multiple second tree top areas; then, the tree crown is depicted based on the multiple second tree top areas and the seasonal change edge image of the multi-temporal remote sensing image to determine the tree crown delineation image. In summary, through the technical solution of the present invention, the multi-scale characteristics and multi-temporal spectral factors of the tree crown can be considered on the basis of the multi-temporal image, so as to more accurately retain independent small tree crowns, improve the accuracy of single tree crown delineation, and provide technical support for the fine identification and classification of tree species.

[0018] The further effects of the above-mentioned non-conventional preferred manner will be described below in conjunction with specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the existing technical solutions, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0020] Figure 1 A schematic diagram of a process flow of a tree crown delineation method provided by an embodiment of the present invention;

[0021] Figure 2 A schematic flow chart of another tree crown delineation method provided by an embodiment of the present invention;

[0022] Figure 3 A schematic diagram of a combination of n first tree top area sets provided by an embodiment of the present invention;

[0023] Figure 4 A schematic structural diagram of a tree crown delineation device provided in an embodiment of the present invention;

[0024] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] like Figure 1 As shown, a method provided by an embodiment of the present invention is provided. The method provided by an embodiment of the present invention can be applied to an electronic device, and can be specifically applied to a server or a general computer. In this embodiment, the method specifically includes the following steps:

[0027] like Figure 1 The embodiment of the present invention provides a tree crown delineation method, comprising the following steps:

[0028] Step 101, determining a multi-temporal remote sensing image of a target forest area; wherein the multi-temporal remote sensing image is an image determined by multiple remote sensing images of the target forest area in different seasons.

[0029] In this step, multi-temporal remote sensing images are introduced into the crown delineation to improve the accuracy of delineating the crowns of two adjacent tree species in forests with complex tree species categories, more accurately retain independent small crowns, and thus improve the overall accuracy of crown delineation.

[0030] Specifically, the target forest area can be understood as an area for large areas of trees, usually a forest area, such as the Amazon forest.

[0031] As a feasible implementation method, multiple remote sensing images of different seasons are all high-resolution images, such as 0.4 resolution, or easily accessible Google remote sensing images, thereby reducing the difficulty of data acquisition while ensuring data accuracy.

[0032] As a feasible implementation method, the number of bands of the multi-temporal remote sensing image is the sum of the number of bands of multiple remote sensing images of the target forest area in different seasons.

[0033] As a feasible implementation method, multiple remote sensing images of different seasons may include remote sensing images of the target forest area before budding in spring, remote sensing images after budding in spring, remote sensing images before leaf fall in autumn, and remote sensing images after leaf fall in autumn. For example, multiple remote sensing images of different seasons are all RGB three-band images. Correspondingly, the bands of the multi-temporal remote sensing image can be 12.

[0034] As a feasible implementation method, multiple remote sensing images of different seasons are preprocessed to obtain multi-temporal remote sensing images. Here, image preprocessing is to facilitate the subsequent single tree crown delineation process.

[0035] Please refer to Figure 2 ,Image preprocessing can include multi-phase image superposition, image enhancement and image brightness extraction.

[0036] For example, for multi-phase image superposition, it can be implemented in the following way:

[0037] Firstly, the remote sensing images with better quality in the spring and autumn phases of the study area (target forest area) are registered to obtain multiple registered image groups; the registered image group with the best quality is determined from the multiple registered image groups, and the four high-resolution phase remote sensing images in the registered image group are subjected to band superposition processing, so that the four three-band RGB high-resolution images are superimposed into a 12-band image MSL.

[0038] It is worth noting that multiple remote sensing images of different seasons can be registered remote sensing images or unregistered remote sensing images. Registration can be understood as the extraction of the same area in different images.

[0039] In one example, image enhancement can be implemented in the following ways:

[0040] An image enhancement operation is performed on the superimposed image MSL to obtain an enhanced image MSS.

[0041] It should be noted that the purpose of image enhancement is to expand the grayscale distribution range of the tree crown in the corresponding band and increase the contrast between the tree crowns. While effectively improving the visual effect of the tree crown image, it also enhances the distinction between the tree crowns, which is conducive to the subsequent extraction of single tree crowns.

[0042] In an example, image brightness extraction can be implemented in the following two ways:

[0043] Implementation method 1: Perform principal component analysis (PCA) on the enhanced image MSS to determine the brightness value of each pixel in the enhanced image MSS to obtain the brightness remote sensing image MSm.

[0044] In practical applications, the first-dimensional principal component (PC1) of the enhanced image MSS after PCA transformation represents the brightness information of the enhanced image MSS, thereby obtaining the brightness remote sensing image MSm.

[0045] Implementation method 2: Perform image averaging processing on the enhanced image MSS to obtain a single-band brightness remote sensing image MSm that can highlight brightness information.

[0046] Please refer to Figure 2 Furthermore, image preprocessing may also include non-vegetation information removal processing.

[0047] In one example, the non-vegetation information removal process can be implemented in the following manner:

[0048] For each pixel in the enhanced image MSS, the vegetation index of the pixel is determined, and based on the comparison result between the vegetation coefficient and the preset vegetation coefficient threshold Thr, it is determined whether to remove the brightness value of the pixel in the brightness remote sensing image MSm; and the image MSa after eliminating the non-vegetation area in the brightness remote sensing image MSm is obtained.

[0049] For example, if the multiple remote sensing images of different seasons are RGB images, the vegetation coefficient may be a Normalized green-red difference index (NGRDI) vegetation index.

[0050] In practical applications, for each pixel in the enhanced image MSS, the NGRDI vegetation index of the first three, middle first three, middle last three or last three bands (i.e., the RGB three bands of a remote sensing image in multiple remote sensing images of different seasons) of the pixel is calculated. The vegetation index is calculated by the following formula (1):

[0051]

[0052] Among them, G represents the DN value (band value) of the green light channel, and R represents the DN value of the red light channel.

[0053] For example, the preset vegetation coefficient threshold Thr can be determined based on the vegetation system of each pixel in the enhanced image MSS. Further, the non-vegetation area outside the threshold is removed from the brightness remote sensing image MSm to obtain an image MSa without the non-vegetation area.

[0054] Please refer to Figure 2 ,Furthermore, the image preprocessing can also include grass information removal processing.

[0055] In one example, the grass information removal process can be implemented in the following way:

[0056] Gray Level Co-occurrence Matrices (GLCM) are used to extract grassland information from the image MSa without vegetation areas, and the grassland information extracted from the image MSa is eliminated to obtain the image MSb without grassland information.

[0057] Since the texture information of trees and grass is different, the grass information can be extracted by using Gray Level Co-occurrence Matrices (GLCM) and then further removed.

[0058] For the image MSa without vegetation, assuming the size is M*N, and the gray levels are s1 and s2, the gray level co-occurrence matrix with a certain spatial relationship is p(s1,s2)

[0059] P(s1, s2)=#{[(r1,c1),(r2,c2)]∈M·N|I(r1,c1)=s1,I(r2,c2)

[0060] =s2}

[0061] In the formula, #{x} represents the number of elements in the set x, (r1, c1) and (r2, c2) represent a point pair in the image space, and r1, c1, r2, and c2 represent the pixel coordinates of two points in the image. Take the pixel pair spacing as d, the direction of the line connecting the two pixels as α, and take the grayscale co-occurrence matrix with different spacings d and different directions α as p(s1, s2, d, α). Take a certain d value, set α to 0°, 45°, 90°, and 135° respectively, to form four grayscale co-occurrence matrices. The superposition of these four directions is used to eliminate the influence of direction. The spatial co-occurrence characteristics of grayscale values ​​are used as a measure of texture. The grayscale space of fine textures changes very quickly, while coarse textures only change slightly with increasing distance. Use different weight matrices to filter the co-occurrence matrix, and select the following statistical properties to describe texture features:

[0062] Angular second moment (Q1): Q1 = Σ s1 Σ s2 (p(s1, s2)) 2 , mainly monitors the degree of global uniformity of the texture;

[0063] Contrast (Q2): Q2 = ∑ s1 ∑ s2 |s1—s2| k p(s1,s2), mainly monitors the image contrast edge and its edge effect;

[0064] Entropy (Q3): Q3 = —∑ s1 ∑ s2p(s1, s2)·lgp(s1, s2), mainly detects the complexity of image space and internal uniformity;

[0065] Correlation (Q4): Q4 = |∑ s1 Σ s2 s1·s2·p(s1,s2)—μ1·μ2| / (σ1·σ2), where μ1, μ2, σ1, σ2 are the mean and variance of p(s1,s2), respectively. This statistic mainly detects the spatial variation geometric structure of the neighborhood mean and variance.

[0066] The gray-level co-occurrence matrix of image MSa is obtained, including angular second-order moment, contrast, entropy and correlation. The thresholds of angular second-order moment, contrast, entropy and correlation of the grass are selected for the four statistical attributes describing the texture features, and the grass information within the threshold is removed to achieve the purpose of removing the grass information and obtain the image MSb without grass information.

[0067] It should be noted that the above image preprocessing is only an example and does not constitute a specific limitation. The specific needs should be determined in combination with actual needs.

[0068] For example, the multi-temporal remote sensing image in this embodiment may be an image MSb from which grassland information is removed.

[0069] Step 102: perform tree top region detection on multiple filtered images of the multi-temporal remote sensing image, and determine multiple first tree top region sets corresponding to the multiple filtered images, respectively; wherein the sizes of filters used by the multiple filtered images are different.

[0070] In this step, multi-scale features and multi-temporal remote sensing images are combined for tree crown delineation, taking into account the changes in different crown diameters and different seasons in the target forest area. Independent small crowns can be retained more accurately, thereby improving the accuracy of single tree crown delineation and providing technical support for the fine identification and classification of tree species.

[0071] Please refer to Figure 2 As a feasible implementation method, multiple first tree top area sets are determined through the following two aspects.

[0072] 1) Filtering

[0073] The multi-temporal remote sensing images are filtered based on multiple filters to remove noise and branches, and the filtered images corresponding to each of the multiple filters are determined.

[0074] Suppose that the diameters of the tree crowns in the multi-temporal remote sensing image, such as the image MSb without grassland information, are {S1, S2, ..., S n}, then: the filter size L = S i(i=1,…,n). Correspondingly, for each of the n crown diameters, the crown diameter is used as the size of the filter, the shape of the filter can be square or circular, and the image MSb is filtered based on the filter to determine the filtered image MS a_Tmp. Here, the n crown diameters correspond to n filters one by one, and the size of each of the n filters is the corresponding tree hanging diameter, and the n filters correspond to the n filtered images MS a_Tmp one by one. It is worth noting that the crown diameter is usually determined by interpolation between the maximum diameter value and the minimum diameter value input manually.

[0075] For example, filtering is performed using the following formula (2):

[0076]

[0077] Where (x, y) represents the coordinates of the pixel in the multi-temporal remote sensing image; σ represents the standard deviation. It should be noted that the standard deviation σ is set according to the size L of the filter, and the empirical value 0.3L is taken as the standard deviation value of the filter.

[0078] 2) Extracting tree tops: Determining initial markers (based on multi-scale shape features)

[0079] For each of the multiple filtered images, the local maximum value of the top of the tree crown area in the filtered image is detected to determine a number of tree top areas, and the multiple tree top areas are taken as a tree top area set. Further, the tree top area sets corresponding to the multiple filtered images are obtained. For the convenience of description and distinction, the tree top area sets corresponding to the multiple filtered images are respectively referred to as first tree top area sets. It should be noted that the first tree top area in the first tree top area set is the initially identified tree top.

[0080] In practical applications, the built-in function imregionalmax in the MATLAB program is used to detect the local maximum region treeTopsCur at the top of the tree canopy, and the obtained multiple local maximum regions treeTopsCur are respectively used as tree top regions. In other words, the first tree top region set includes several local maximum regions treeTopsCur obtained by performing local maximum detection on the image MSb after the grass information is removed by the corresponding filter. Figure 3 shows n first tree top area sets LS1 to LS n .

[0081] Furthermore, considering that there may be non-tree top areas in the first tree top area set, in order to ensure the accuracy of the tree top areas, for each of the multiple first tree top area sets, a non-tree top area judgment is performed on each first tree top area set in the first tree top area set, and the first tree top areas judged as non-tree top areas are deleted.

[0082] Optionally, the non-tree top region may be judged by roundness. For example, if the roundness of the tree top region is greater than or equal to a roundness threshold, it is retained, otherwise it is deleted.

[0083] The circularity is a measure of the similarity between the area of ​​a line segment and a circle. For example, it can be calculated by the following formula (3):

[0084]

[0085] Among them, c represents the circularity (0<c<1), A is the area of ​​the measurement region (local maximum region treeTopsCur), and r is the radius of the measurement region (local maximum region treeTopsCur).

[0086] For example, suppose Figure 3 If the roundness of the tree top region numbered 21 in the first tree top region set LS2 is greater than the roundness threshold, it is deleted.

[0087] Step 103: Based on the multi-temporal spectral information of the multi-temporal remote sensing image, multiple first tree top regions in the multiple first tree top region sets are screened and merged to determine multiple second tree top regions.

[0088] In this step, the multi-temporal spectral information of multi-temporal remote sensing images is used to screen and merge multiple tree top areas, which can fully consider the seasonal changes between different crown areas and more accurately retain independent small crowns, thereby improving the accuracy of single tree crown delineation and providing technical support for the fine identification and classification of tree species.

[0089] It is worth noting that the multi-temporal spectral information of the multi-temporal remote sensing image is used to represent the differences between multiple remote sensing images of different seasons. By way of example, it may include the spectral information of each of the multiple remote sensing images of different seasons, and may also include the spectral information of each of the multiple images with better performance among the multiple remote sensing images of different seasons. The images with better performance can be understood as images that can better characterize the differences between the tree crowns of the target forest area. By way of example, the multi-temporal spectral information of the multi-temporal remote sensing image includes the band values ​​(R value, G value, B value) of each of the four remote sensing images (RGB) of spring and autumn, that is, the values ​​of each of the multiple color channels.

[0090] As a feasible implementation method, the multi-temporal spectral information of the multi-temporal remote sensing image is used to determine the seasonal change information between multiple overlapping first tree top areas in any multiple sets of multiple first tree top area sets, and the seasonal change information is used to determine whether to combine the multiple overlapping first tree top areas into one tree top area.

[0091] It should be noted that the combination of multiple overlapping tree top regions can be understood as: multiple tree top regions that overlap partially or completely. For example, for any two sets of multiple first tree top region sets, if the tree top region i in one set intersects with the n tree top regions in another set and includes the m tree top regions in the other set, then the tree top region i, the n tree top regions, and the m tree top regions are multiple overlapping tree top regions. For example, Figure 3 The tree top regions numbered 22, 12, and 13 in are overlapped, and the tree top regions numbered 21 and 11 are also overlapped.

[0092] It is worth noting that combining the overlapping multiple first tree top regions into one tree top region can be understood as a closed region formed by the outer edges of the overlapping multiple first tree top regions. Obviously, the combined tree top region includes the overlapping multiple first tree top regions. For example, Figure 3 The tree top regions labeled 11 and 21 in , since 21 completely covers 11, we can get the tree top region labeled 21. Figure 3 By combining the tree top regions numbered 12, 13, and 22 in , a new region can be obtained, which is a closed region formed by the outer edges of the overlapping tree top regions 12, 13, and 22, including the tree top regions numbered 12, 13, and 22.

[0093] For example, when the seasonal change information indicates that the seasonal change between the overlapping multiple first tree top regions is small, the overlapping multiple first tree top regions are combined into one tree top region; otherwise, some of the overlapping multiple first tree top regions are selected.

[0094] In an example, the plurality of second tree-top regions include a region obtained by merging a plurality of overlapping first tree-top regions in a plurality of first tree-top region sets and / or a partial region of a plurality of overlapping first tree-top regions in a plurality of first tree-top region sets.

[0095] For ease of description, the overlapping multiple first tree top regions in the multiple first tree top region sets are taken as overlapping sets, and the multiple first tree top region sets have multiple overlapping sets. Correspondingly, the multiple second tree top regions include the regions after the multiple first tree top regions in the partial sets in the multiple overlapping sets are merged, and the partial first tree top regions in the other partial sets. The partial first tree top regions in the overlapping sets can be understood as the first tree top regions in the overlapping sets that are smaller than the preset region area, or can also be understood as several first tree top regions with the smallest filter size in the overlapping sets.

[0096] Furthermore, the plurality of second tree-top regions also include a plurality of non-overlapping first tree-top regions in the plurality of first tree-top region sets.

[0097] As a feasible implementation, multiple second tree top areas may be determined in the following implementation:

[0098] According to the order of filter sizes from small to large, multiple first tree top area sets are sorted to determine a set sequence; for the i-th first tree top area set in the set sequence, based on the multi-temporal spectral information of the multi-temporal remote sensing image, the i-2-th second tree top area set and the i-th first tree top area set are combined to determine the i-1-th second tree top area set; wherein, when i=3, the first second tree top area set is obtained by combining the two first tree top area sets that are located at the first position in the set sequence; each tree top area in the second tree top area set determined by the first tree top area set that is located at the last position in the set sequence is respectively used as the second tree top area.

[0099] like Figure 3 As shown, the first tree top area set of each filtered image is used as a layer of LS in the horizontal slice i (i=1, ..., n), the obtained multi-layer slices construct the tree top slices {LS1, LS2, ..., LS n}, the tree top slices are sorted from small to large according to the crown diameter (filter size), and the tree top slices {LS1, LS2, ..., LS n} is integrated, that is, two adjacent first tree top area sets in multiple first tree top area sets are combined to obtain the final layer uLS n-1 It should be understood that the final layer uLS n Includes the top slices {LS1, LS2, ..., LS n The first tree top area set corresponding to each layer in} can also be understood as a set sequence.

[0100] Specifically, Figure 3As shown, for the combination process, starting from the first layer seed slice, the first layer slice LS1 and the second layer slice LS2 are combined to form a combined layer uLS1; then the new combined layer uLS1 is combined with the third layer LS3 to form a new combined layer uLS2; then the new combined layer uLS i With its upward adjacent layer LS i+2 Combine to form a new combined layer uLS i+1 Repeat this combination process until the LS n layer, and obtain the combined layer uLS of the layer with the largest crown diameter n-1 , thus obtaining the combined layer treetopsT. Correspondingly, the combined layer uLS of the layer with the largest crown diameter n-1 Each tree top region in is used as the second tree top region. Here, uLS i That is, the above-mentioned i-2 second tree top area set, uLS i and LS i+2 Combine to determine uLS i+1 .

[0101] It should be noted that the combination of the i-2th second tree top region set and the i-th first tree top region set can be understood as Figure 2 In one example, the combination of the i-2th second tree top region set and the i-th first tree top region set can be implemented in the following manner:

[0102] For multiple overlapping tree top areas in the i-2th second tree top area set and the i-th first tree top area set, seasonal change information is determined based on multi-temporal spectral information of multi-temporal remote sensing images, and the seasonal change information indicates seasonal changes between the multiple tree top areas; based on the seasonal change information, the multiple tree top areas are combined into a tree top area, or the multiple tree top areas are added to the i-1th second tree top area set; the multiple non-overlapping tree top areas in the i-2th second tree top area set and the i-th first tree top area set are added to the i-1th second tree top area set.

[0103] Specifically, Figure 3 As shown, for the first uLS i Layer and LS i+1 For overlapping tree top regions in a layer, if the roundness of the tree top region is greater than or equal to the roundness threshold, it is retained, otherwise it is deleted. For example, delete Figure 3 The top area of ​​the tree with the label 22; then, the uLS i Layer and LS i+1 Multiple tree top areas that do not overlap at each layer are added to the uLS i+1 Layer; for the uLS i Layer and LSi+1 If there are multiple overlapping tree top areas, and the seasonal changes between these tree top areas are small, these tree top areas are merged into one tree top area and added to the first uLS i+1 Otherwise, add some of the tree top areas in these tree top areas to the first uLS i+1 layer.

[0104] Optionally, the non-tree top area can be determined by roundness. See above for details.

[0105] For example, the seasonal change information includes several seasonal change values, and the seasonal change information can be determined by the following implementation method:

[0106] Based on the multi-temporal spectral information of multiple tree top areas corresponding to the multi-temporal remote sensing images, a reference spectrum is determined; for each tree top area located in the i-2th first tree top area set, a target spectrum is determined based on the multi-temporal spectral information of the tree top area corresponding to the multi-temporal remote sensing images, and based on the target spectrum and the reference spectrum, the seasonal change value corresponding to the tree top area is determined.

[0107] Correspondingly, when each seasonal change value in the seasonal change information is greater than the seasonal change threshold, a tree top area composed of multiple tree top areas is added to the i-1th second tree top area set; otherwise, at least one tree top located in the i-2th first tree top area set is added to the i-1th second tree top area set.

[0108] It should be noted that the multi-temporal spectral information of the tree top area corresponding to the multi-temporal remote sensing image can be understood as the multi-temporal spectral information of the position of the tree top area in the multi-temporal remote sensing image, that is, the multi-temporal spectral information of the area where the tree top area is projected onto the multi-temporal remote sensing image. It should be understood that the tree top area has position information, which is used to indicate its position in the multi-temporal remote sensing image.

[0109] For example, the seasonal phase change threshold is determined based on seasonal phase change values ​​between a plurality of overlapping tree top regions in two adjacent first tree top region sets in the set sequence.

[0110] The seasonal phase change value can be expressed by the cosine value of the spectral angle cos(θ). The larger the cosine value, the greater the seasonal phase change between the two tree crowns and the greater the difference. Specifically, the cosine value cos(θ) can be calculated using the following formula (4):

[0111]

[0112] Among them, a is the canopy pixel spectrum to be compared, b is the reference canopy spectrum, and n is the number of image bands.

[0113] In practical applications, the cosine values ​​cos(θ) between multiple overlapping treetop areas of two adjacent first treetop area sets in the set sequence are calculated in turn; then, all the calculation results of two adjacent first treetop area sets in the set sequence are counted and the mode is taken, and the mode is used as the rough value of the threshold; then, the threshold is fine-tuned in a small range near the rough value, and the final spectral threshold sam (seasonal change value) is determined according to the actual extraction effect.

[0114] For example, for the convenience of description and distinction, the i-2th second treetop region set is called a fine layer, and the i-th first treetop region set is called a coarse layer; that is, the uLS i The LS layer is a fine layer, i.e., a layer with a smaller filter size. i+1 The layer is a coarse layer, i.e., a layer with a larger filter size. The set formed by combining the i-2th second treetop region set and the i-th first treetop region set is called a pre-combined layer. It should be understood that the pre-combined layer is a set formed by merging the i-2th second treetop region set and the i-th first treetop region set after eliminating non-treetop regions, i.e., retaining non-overlapping treetop regions and merging overlapping multiple treetop regions.

[0115] The i-1th second tree top region set can be determined by the following implementation method:

[0116] a) For one of the multiple tree top regions in the pre-combined layer, determine the position of the tree top region, and find the corresponding positioning regions of the fine layer and the coarse layer at the same position according to the position. Here, the positioning region refers to the region with the same shape as the tree top region.

[0117] b) Determine whether there is a tree top area in the positioning area corresponding to the fine layer and the coarse layer. If only one layer has a tree top area in the positioning area, output the tree top area of ​​the layer. If both layers have tree tops in their corresponding areas, proceed to 3).

[0118] c) The average spectrum of all pixels in the multi-temporal remote sensing image corresponding to the positioning area corresponding to the rough layer is used as the reference spectrum b. The average spectrum includes the average of the band values ​​of the multiple pixels in the multiple bands.

[0119] d) Determine the number m of sub-regions (tree top regions) in the corresponding positioning area in the fine layer. For each sub-region, the average spectrum of all pixels corresponding to the sub-region in the multi-temporal remote sensing image is used as the target spectrum of the sub-region, forming a total of m target spectra a.

[0120] e) According to the above formula (4), the cosine values ​​of the spectral angles between the m target spectra and the reference spectrum are calculated to obtain the SAM calculation result of each sub-region.

[0121] f) Integrate the treetops on the coarse layer and the fine layer according to the set spectral threshold. If the SAM calculation results of each sub-region are greater than the threshold sam, it means that these sub-regions in this position on the fine layer have similar spectra and small seasonal changes, and are integrated into one treetop region; if the calculation result is less than the threshold sam, it means that the spectra of these sub-regions in this position on the fine layer are very different and have large seasonal changes, and the m sub-regions in the fine layer are retained as independent treetop regions.

[0122] g) traverse and locate all tree top regions in the pre-combined layer in sequence, repeat the above operations a) to f) for each tree top region to obtain the final tree top region, and form the i-1th second tree top region set.

[0123] After the above determination, the tree top region in the second tree top region set finally determined is used as the second tree top region. Figure 3 uLS n-1 Each tree top region in the layer serves as a second tree top region.

[0124] Step 104: Delineate the tree crown based on the plurality of second tree top regions and the seasonal variation edge images of the multi-temporal remote sensing images to determine the tree crown delineation image.

[0125] In one example, the tree crown delineation image can be determined in the following implementation method:

[0126] A foreground image is determined based on multiple second treetop areas; a seasonal change edge image of a multi-temporal remote sensing image is used as a background image; and a crown segmentation is performed based on the foreground image and the background image using a watershed algorithm to determine a crown delineation image.

[0127] Specifically, tree crown depiction can be achieved through the following two aspects.

[0128] 1) Extract edge information of seasonal changes.

[0129] Specifically, the following five methods for extracting edge information of seasonal changes are selected: Laplace image sharpening, LOG (Laplace and Guassian) operator, Sobel edge operator, Gradient gradient function, and Prewitte operator.

[0130] The MSS image is extracted by five extraction methods in turn to obtain the seasonal change edge image MSpj (j=1,…,5). The seasonal change edge image with the best effect is selected as the background image.

[0131] It should be noted that the extracted seasonal change edge information can be obtained in advance during the image preprocessing process in step 101 .

[0132] 2) Watershed segmentation

[0133] This embodiment uses the seasonal change edge image as the background image, uses images determined based on multiple second tree top areas as the marker image, performs tree crown segmentation through marker watershed segmentation, and obtains the final tree crown delineation image segMap.

[0134] The size of the marked image and the multi-temporal remote sensing image is consistent, and the marked image can be a binary image. For example, the multiple second tree top regions can be black (the background is white) or white (the background is black). For example, the image obtained by marking the multiple second tree top regions on the multi-temporal remote sensing image is binarized to obtain the marked image. For example, the image obtained by marking the multiple second tree top regions on the multi-temporal remote sensing image is binarized to obtain the marked image.

[0135] It should be understood that the calculation process of the watershed is an iterative labeling process. From the principle of the watershed segmentation algorithm, it can be seen that: the segmentation algorithm mainly detects the edges of the image, and then segments the target according to the edge ridges of the target, and finally obtains the segmentation line image of the water collection basin. In view of the fact that the seasonal change edge image represents the edge features between the target and the background, it is beneficial to the edge detection of the target image. In this embodiment, the grayscale value on the seasonal change edge image is analogous to the water level value, and multiple second tree top areas are used as minimum values ​​to construct a water collection basin.

[0136] like Figure 2 As shown, further, the segmentation results need to be optimized.

[0137] As a feasible implementation method, the optimization method includes removing isolated point noise and removing holes.

[0138] 1) Remove isolated point noise

[0139] Calculate the area of ​​the smallest crown segmentation patch in the crown delineation image segMap, and set this area as Sa. Set Sa as the area threshold for judging whether the segmentation patch is an isolated point noise. Patches with an area smaller than this value are regarded as isolated point noise and removed. Retain the segmentation patches with an area larger than this value and re-compose a new crown segmentation map treeLayers.

[0140] 2) Remove holes

[0141] First, the crown segments in treeLayers are marked with regional attributes, and the corresponding position index and a logical matrix of the same size and shape are established; then, the segments are traversed according to the index and the regions are filled; finally, the filled binary image is output to obtain the crown segmentation map tmp after removing holes, which is the final result of single tree crown delineation.

[0142] Through the above technical solution, it can be seen that the beneficial effects of this embodiment are:

[0143] Based on the multi-temporal images, the multi-scale characteristics and multi-temporal spectral factors of the tree crown are taken into consideration. The tree crowns are screened and merged based on seasonal changes to more accurately retain independent small tree crowns, thereby improving the accuracy of single tree crown delineation and providing technical support for the fine identification and classification of tree species.

[0144] Based on the same concept as the embodiment of the method of the present invention, please refer to Figure 4 The embodiment of the present invention further provides a tree crown delineation device, comprising:

[0145] An image determination unit 401 is used to determine a multi-temporal remote sensing image of a target forest area; wherein the multi-temporal remote sensing image is an image determined based on multiple remote sensing images of the target forest area in different seasons;

[0146] The tree top detection unit 402 is used to perform tree top area detection on the multiple filtered images of the multi-temporal remote sensing image, and determine multiple first tree top area sets corresponding to the multiple filtered images respectively; wherein the filter sizes used by the multiple filtered images are different;

[0147] A screening unit 403 is configured to screen and merge a plurality of first tree top regions in the plurality of first tree top region sets based on the multi-temporal spectral information of the multi-temporal remote sensing image to determine a plurality of second tree top regions;

[0148] The tree crown delineation unit 404 is used to perform tree crown delineation based on the multiple second tree top areas and the seasonal phase change edge image of the multi-temporal remote sensing image to determine a tree crown delineation image.

[0149] As a feasible implementation method, the multiple remote sensing images of different seasons include: remote sensing images of the target forest area before spring budding, remote sensing images after spring budding, remote sensing images before autumn leaf fall, and remote sensing images after autumn leaf fall.

[0150] As a feasible implementation method, the multiple remote sensing images of different seasons are all RGB images.

[0151] As a feasible implementation method, the multi-temporal remote sensing image is obtained based on image preprocessing of multiple remote sensing images of the target forest area in different seasons, and the image preprocessing includes image enhancement and image brightness extraction.

[0152] As a feasible implementation manner, the image preprocessing further includes non-vegetation information removal processing and / or grassland information removal processing.

[0153] As a feasible implementation method, the multi-temporal spectral information of the multi-temporal remote sensing image is used to determine the seasonal change information between the overlapping multiple first tree top areas in any multiple sets of the multiple first tree top area sets, and the seasonal change information is used to determine whether to combine the overlapping multiple first tree top areas into one tree top area.

[0154] As a feasible implementation manner, the multiple second tree-top regions include a merged region of multiple overlapping first tree-top regions in the multiple first tree-top region sets and / or a partial region of multiple overlapping first tree-top regions in the multiple first tree-top region sets.

[0155] As a feasible implementation manner, the multiple second tree-top regions further include multiple non-overlapping first tree-top regions in the multiple first tree-top region sets.

[0156] As a feasible implementation, the screening unit 403 includes: a sorting unit, a combining unit and a region determining unit; wherein,

[0157] The sorting unit is used to sort the plurality of first tree top region sets in the order of the size of the filter from small to large, so as to determine a set sequence;

[0158] The combining unit is used for combining the i-2th second tree top area set and the i-th first tree top area set for the i-th first tree top area set of the set sequence based on the multi-temporal spectral information of the multi-temporal remote sensing image to determine the i-1th second tree top area set; wherein when i=3, the first second tree top area set is obtained by combining the two first tree top area sets located at the first positions in the set sequence;

[0159] The region determination unit is configured to use each tree top region in a second tree top region set determined by a first tree top region set located at the last position in the set sequence as a second tree top region.

[0160] As a feasible implementation, the combination unit includes: a first combination subunit and a second combination subunit; wherein,

[0161] The first combining subunit is used to determine, for a plurality of overlapping tree top regions in the (i-2)th second tree top region set and the (i)th first tree top region set, seasonal phase change information based on multi-temporal spectral information of the multi-temporal remote sensing image, wherein the seasonal phase change information indicates seasonal phase changes between the plurality of tree top regions; based on the seasonal phase change information, combine the plurality of tree top regions into a tree top region, or add the plurality of tree top regions to the (i-1)th second tree top region set;

[0162] The second combining subunit is used to add a plurality of non-overlapping tree top regions in the (i-2)th second tree top region set and the (i)th first tree top region set to the (i-1)th second tree top region set.

[0163] As a feasible implementation method, the seasonal change information includes at least one seasonal change value; the first combination subunit is used to determine a reference spectrum based on the multi-temporal spectral information of the multiple tree top areas corresponding to the multi-temporal remote sensing image; for each tree top area located in the i-2th first tree top area set, a target spectrum is determined based on the multi-temporal spectral information of the tree top area corresponding to the multi-temporal remote sensing image, and the seasonal change value corresponding to the tree top area is determined based on the target spectrum and the reference spectrum.

[0164] As a feasible implementation method, the first combination subunit is used to determine the seasonal change threshold based on the seasonal change values ​​between multiple overlapping tree top areas in two adjacent first tree top area sets in the set sequence; when each seasonal change value in the seasonal change information is greater than the seasonal change threshold, a tree top area formed by combining the multiple tree top areas is added to the i-1th second tree top area set; otherwise, at least one tree top located in the i-2th first tree top area set is added to the i-1th second tree top area set.

[0165] As a feasible implementation, the tree crown depiction unit 304 includes: an image determination unit and a segmentation unit; wherein,

[0166] The image determination unit is used to determine a foreground image based on the multiple second tree top regions; and use the seasonal phase change edge image of the multi-temporal remote sensing image as a background image;

[0167] The segmentation unit is used to perform tree crown segmentation based on the foreground image and the background image by using a watershed algorithm to determine a tree crown delineation image.

[0168] As a feasible implementation method, the device further includes: an optimization module: wherein,

[0169] The optimization module is used to sequentially remove isolated point noise and holes in the tree crown circle image to determine a tree crown delineation result image.

[0170] Figure 5It is a structural diagram of an electronic device provided by an embodiment of the present invention. At the hardware level, the electronic device includes a processor 501 and a memory 502 storing execution instructions, and optionally also includes an internal bus 503 and a network interface 504. Among them, the memory 502 may include a memory 5021, such as a high-speed random access memory (Random-Access Memory, RAM), and may also include a non-volatile memory 5022 (non-volatile memory), such as at least one disk storage, etc.; the processor 501, the network interface 504 and the memory 502 can be interconnected through an internal bus 503, and the internal bus 503 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc.; the internal bus 503 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one bidirectional arrow is used in the figure, but it does not mean that there is only one bus or one type of bus. Of course, the electronic device may also include hardware required for other services. When the processor 501 executes the execution instruction stored in the memory 502, the processor 501 executes the method in any embodiment of the present invention, and is at least used to execute the following steps: Figure 1 or Figure 2 The method shown.

[0171] In a possible implementation, the processor reads the corresponding execution instructions from the non-volatile memory into the memory and then runs them, and can also obtain the corresponding execution instructions from other devices to form a tree crown delineation device at the logical level. The processor executes the execution instructions stored in the memory to implement a tree crown delineation method provided in any embodiment of the present invention through the execution of the execution instructions.

[0172] The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or instructions in the form of software. The above processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present invention can be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0173] The embodiment of the present invention further provides a computer-readable storage medium, including an execution instruction. When a processor of an electronic device executes the execution instruction, the processor executes the method provided in any embodiment of the present invention. The electronic device may be specifically as follows: Figure 5 The electronic device shown; the execution instruction is a computer program corresponding to the crown delineation device.

[0174] Those skilled in the art will appreciate that the embodiments of the present invention may be provided as methods or computer program products. Therefore, the present invention may be implemented in the form of a complete hardware embodiment, a complete software embodiment, or a combination of software and hardware.

[0175] The various embodiments of the present invention are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0176] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0177] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A tree crown delineation method, characterized in that: include: Determine a multi-temporal remote sensing image of a target forest area; wherein the multi-temporal remote sensing image is an image determined based on multiple remote sensing images of the target forest area in different seasons; Performing tree top region detection on a plurality of filtered images of the multi-temporal remote sensing image, respectively, to determine a plurality of first tree top region sets corresponding to the plurality of filtered images, respectively; wherein the sizes of filters used by the plurality of filtered images are different; Based on the multi-temporal spectral information of the multi-temporal remote sensing image, multiple first tree top areas in the multiple first tree top area sets are screened and merged to determine multiple second tree top areas; the multi-temporal spectral information of the multi-temporal remote sensing image is used to determine the seasonal change information between the multiple overlapping first tree top areas in any multiple sets of the multiple first tree top area sets, and the seasonal change information is used to determine whether to combine the multiple overlapping first tree top areas into one tree top area; the multiple second tree top areas include the area after the overlapping multiple first tree top areas in the multiple first tree top area sets are merged and / or part of the overlapping multiple first tree top areas in the multiple first tree top area sets; the multiple second tree top areas also include the multiple non-overlapping first tree top areas in the multiple first tree top area sets; The tree crown is delineated based on the multiple second tree top areas and the seasonal variation edge image of the multi-temporal remote sensing image to determine the tree crown delineation image.

2. The method according to claim 1, characterized in that The multiple remote sensing images of different seasons include: remote sensing images of the target forest area before budding in spring, remote sensing images after budding in spring, remote sensing images before leaf fall in autumn, and remote sensing images after leaf fall in autumn; The multiple remote sensing images in different seasons are all RGB images; The multi-temporal spectral information of the multi-temporal remote sensing image includes the spectral information of each of the multiple remote sensing images of the target forest area in different seasons; and / or, The multi-temporal remote sensing image is obtained based on image preprocessing of multiple remote sensing images of the target forest area in different seasons; wherein, the image preprocessing includes image fusion, image enhancement and image brightness extraction, and the image preprocessing also includes non-vegetation information elimination processing and / or grassland information elimination processing.

3. The method according to claim 1, characterized in that The method of screening and merging a plurality of first tree top regions in the plurality of first tree top region sets based on the multi-temporal spectral information of the multi-temporal remote sensing image to determine a plurality of second tree top regions includes: Sorting the plurality of first tree top region sets in order of size of the filters from small to large to determine a set sequence; For the i-th first tree top area set of the set sequence, based on the multi-temporal spectral information of the multi-temporal remote sensing image, the i-2-th second tree top area set and the i-th first tree top area set are combined to determine the i-1-th second tree top area set; wherein, when i=3, the first second tree top area set is obtained by combining the two first tree top area sets located at the first positions in the set sequence; Each tree top region in the second tree top region set determined by the first tree top region set located at the last position in the set sequence is used as a second tree top region.

4. The method according to claim 3, characterized in that The combining the (i-2)th second tree top area set and the (i-1)th first tree top area set based on the multi-temporal spectral information of the multi-temporal remote sensing image to determine the (i-1)th second tree top area set includes: For a plurality of overlapping tree top regions in the (i-2)th second tree top region set and the (i)th first tree top region set, seasonal phase change information is determined based on multi-temporal spectral information of the multi-temporal remote sensing image, wherein the seasonal phase change information indicates seasonal phase changes between the plurality of tree top regions; based on the seasonal phase change information, the plurality of tree top regions are combined into a tree top region, or the plurality of tree top regions are added to the (i-1)th second tree top region set; Add the non-overlapping treetop regions in the (i-2)th second treetop region set and the (i)th first treetop region set to the (i-1)th second treetop region set.

5. The method according to claim 4, characterized in that The seasonal phase change information includes at least one seasonal phase change value; The determining of seasonal variation information based on the multi-temporal spectral information of the multi-temporal remote sensing image includes: Determine a reference spectrum based on multi-temporal spectral information of the plurality of tree top regions corresponding to the multi-temporal remote sensing image; For each tree top region located in the (i-2)th first tree top region set, determining a target spectrum based on multi-temporal spectrum information corresponding to the tree top region in the multi-temporal remote sensing image, and determining a seasonal phase change value corresponding to the tree top region based on the target spectrum and the reference spectrum; The combining the plurality of tree top regions into a tree top region based on the seasonal phase change information, or adding the plurality of tree top regions to the (i-1)th second tree top region set, comprises: Determining a seasonal phase change threshold based on seasonal phase change values ​​between a plurality of overlapping tree top regions in two adjacent first tree top region sets in the set sequence; When each seasonal change value in the seasonal change information is respectively greater than the seasonal change threshold, a tree top area formed by combining the multiple tree top areas is added to the i-1th second tree top area set; otherwise, at least one tree top located in the i-2th first tree top area set is added to the i-1th second tree top area set.

6. The method according to claim 1, characterized in that The step of performing tree crown delineation based on the plurality of second tree top areas and the seasonal phase change edge image of the multi-temporal remote sensing image to determine the tree crown delineation image comprises: determining a foreground image based on the plurality of second tree top regions; Using the seasonal phase change edge image of the multi-temporal remote sensing image as a background image; Based on the foreground image and the background image, tree crown segmentation is performed using a watershed algorithm to determine a tree crown delineation image.

7. The method according to claim 1, characterized in that Also includes: The isolated point noise and holes in the tree crown circle image are removed in sequence to determine the tree crown delineation result image.

8. A tree crown delineation device, characterized in that: include: An image determination unit, used to determine a multi-temporal remote sensing image of a target forest area; wherein the multi-temporal remote sensing image is an image determined based on a plurality of remote sensing images of the target forest area in different seasons; A tree top detection unit is used to perform tree top area detection on a plurality of filtered images of the multi-temporal remote sensing image, and determine a plurality of first tree top area sets corresponding to the plurality of filtered images, respectively; wherein the sizes of filters used by the plurality of filtered images are different; A screening unit, for screening and merging a plurality of first tree top areas in the plurality of first tree top area sets based on the multi-temporal spectral information of the multi-temporal remote sensing image, to determine a plurality of second tree top areas; the multi-temporal spectral information of the multi-temporal remote sensing image is used to determine seasonal variation information between a plurality of overlapping first tree top areas in any plurality of sets of the plurality of first tree top area sets, the seasonal variation information being used to determine whether to combine the plurality of overlapping first tree top areas into one tree top area; the plurality of second tree top areas include an area after merging a plurality of overlapping first tree top areas in the plurality of first tree top area sets and / or a partial area of ​​a plurality of overlapping first tree top areas in the plurality of first tree top area sets; the plurality of second tree top areas also include a plurality of non-overlapping first tree top areas in the plurality of first tree top area sets; A tree crown depiction unit is used to perform tree crown depiction based on the multiple second tree top areas and the seasonal phase change edge image of the multi-temporal remote sensing image to determine a tree crown delineation image.

Images