Visible light image vegetation information extraction method

Abstract

The invention discloses a visible light image vegetation information extraction method, and relates to the technical field of satellite remote sensing image processing, and the method comprises the steps: obtaining a visible light image of a vegetation to-be-detected region; segmenting the visible light image by adopting a multi-scale segmentation algorithm, determining segmentation scales of different vegetations, and determining classification levels of the vegetations according to the segmentation scales of the different vegetations; extracting feature information on each segmentation scale, selecting vegetation features by adopting a random forest according to the extracted feature information, and identifying the selected vegetation features to obtain an identification result; and combining the identification results according to the classification level of the vegetation, and determining the spatial distribution of the vegetation. According to the invention, by determining the segmentation scale and the classification level of each vegetation, the spatial distribution of different vegetation can be accurately obtained.

Description

Technical Field

[0001] The invention relates to the field of satellite remote sensing image processing, and in particular to a method for extracting vegetation information from visible light images. Background Art

[0002] Accurately identifying vegetation in mining areas and understanding its spatial distribution is crucial for investigating vegetation diversity, community structure, post-mining land reclamation evaluation, and vegetation recovery status assessment. Remote sensing technology has become the main means of vegetation monitoring in mining areas due to its fast, real-time, and large-scale data acquisition capabilities.

[0003] UAV remote sensing has high flexibility, short image acquisition cycle, and little weather interference, providing a reliable data acquisition method for high-precision and rapid monitoring of vegetation in mining areas. UAVs can be equipped with visible light lenses, and their images have high spatial resolution and rich texture and geometric information. However, current studies mostly use a single segmentation scale, without considering the scale differences between different objects, which makes it impossible to accurately obtain the spatial distribution of different vegetation. Summary of the invention

[0004] The present invention provides a method for extracting vegetation information from a visible light image, which is used to solve the above-mentioned problem existing in the prior art, that is, the problem that it is difficult to accurately obtain the spatial distribution of different vegetation in the prior art. The present invention provides a method for extracting vegetation information from a visible light image, which comprises:

[0005] Acquire a visible light image of the vegetation test area;

[0006] Use multi-scale segmentation algorithm to segment visible light images, determine the best segmentation scale for different vegetation, and determine the classification level of vegetation based on the best segmentation scale for different vegetation;

[0007] Extracting feature information at each segmentation scale, and selecting vegetation features using random forests based on the extracted feature information, identifying the selected vegetation features, and obtaining identification results; wherein the feature information includes spectral features, vegetation indexes, and texture features;

[0008] The identification results are merged according to the classification level of vegetation to determine the spatial distribution of different vegetation.

[0009] Optionally, the identifying the selected vegetation features to obtain the identification results specifically includes:

[0010] The support vector machine is used to identify and extract the selected vegetation features to obtain the recognition results.

[0011] Optionally, the spectral characteristics specifically include:

[0012] The mean of each channel of the RGB image and the HSV image; the HSV image is obtained by converting the visible light image to the HSV image.

[0013] Optionally, the vegetation index specifically includes:

[0014] Excess greenness index EXG, difference enhanced vegetation index DEVI, difference vegetation index VDVI, normalized green-red difference index NGRDI, normalized green-blue difference index NGBDI and red-green ratio index RGRI.

[0015] Optionally, the texture features specifically include:

[0016] The mean, variance, contrast, difference, information entropy, correlation, homogeneity and second-order moment of each channel of RGB images and HSV images.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a method for extracting vegetation information from visible light images. The method obtains a visible light image of the area to be tested, takes into account the differences in spatial distribution scales between different vegetation, and uses a multi-scale segmentation algorithm to divide it into multiple images with different scales, and determines the segmentation scales of different vegetation; according to the segmentation scales of different vegetation, the classification level of vegetation is determined, and spectral features, vegetation index and texture features are extracted at each scale. Taking into account the differences in feature discrimination between different landforms, random forests are used to select vegetation features to obtain the optimal feature combination, and a support vector machine is used to identify the selected vegetation features. By merging the identification results, accurate acquisition of the spatial distribution of vegetation in the mining area is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0019] Figure 1 A flowchart of a method for extracting vegetation information from visible light images provided by an embodiment of the present invention.

[0020] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0021] 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 the drawings of the present invention. Obviously, the described embodiments are 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.

[0022] At present, scale and features are two key factors affecting the accuracy of object-oriented vegetation recognition. However, in current studies, different landforms mostly use a unified segmentation scale. The mining area has vegetation such as field crops, trees, shrubs and herbs, with high spatial heterogeneity. The optimal segmentation scales corresponding to different types of vegetation are also different. Therefore, it is difficult to express complex landforms with a single scale. In addition, different levels correspond to different types of landforms, and their optimal recognition features are also different. However, current studies have not considered the differences in the importance of features at different levels, which affects the accuracy of obtaining the spatial distribution of different vegetation.

[0023] In view of the above problems, the present invention provides a method for extracting vegetation information from visible light images. The method performs multi-level and multi-scale segmentation on the vegetation in the mining area, takes into account the differences in spatial distribution scales between different vegetation, and adopts a multi-scale segmentation algorithm to divide it into multiple images with different scales, and determines the optimal segmentation scales of different vegetation; according to the optimal segmentation scales of different vegetation, the classification level of vegetation is determined, and spectral features, vegetation index and texture features are extracted at each scale. Taking into account the differences in feature discrimination between different landforms, random forests are used to select vegetation features to obtain the optimal feature combination; support vector machines are used to identify the selected vegetation features, and the identification results are merged to accurately obtain the spatial distribution of vegetation in the mining area.

[0024] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0025] Figure 1 is a flow chart of a method for extracting vegetation information from a visible light image provided by an embodiment of the present invention. Figure 1 As shown, this embodiment shows a method for extracting vegetation information from a visible light image, comprising:

[0026] S101: Acquire a visible light image of the vegetation test area.

[0027] Exemplarily, UAV visible light aerial photography is adopted to obtain centimeter-level UAV visible light images, which are set as aerial photography data; and the aerial photography data is preprocessed to obtain orthophotod UAV visible light images; and the HSV image can be generated by performing HSV image conversion on the visible light image.

[0028] Exemplarily, the visible light image in this embodiment includes ground objects such as corn, sunflower, alfalfa, larch, salix psammophila, wormwood, grassland, and bare soil.

[0029] S102: Use a multi-scale segmentation algorithm to segment the visible light image, determine the best segmentation scale for different vegetation, and determine the classification level of vegetation based on the best segmentation scale for different vegetation.

[0030] Optionally, determining different vegetation segmentation scales includes segmenting the visible light image using a multi-scale segmentation algorithm based on the visible light image to determine an optimal segmentation scale.

[0031] Among them, corn, sunflower and alfalfa are distributed in a large area, so the optimal segmentation scale is larger. The optimal segmentation scale parameters are determined to be 210, the shape factor is 0.1, and the compactness is 0.5; the optimal segmentation scale of Pinus sylvestris, Salix psammophila and bare soil is 180, the shape factor is 0.1, and the compactness is 0.5; Artemisia ordosica is distributed in clusters and is easily mixed with grassland, so the minimum segmentation scale of the two is 70, the shape factor is 0.1, and the compactness is 0.5.

[0032] Optionally, the classification level includes a first classification level, a second classification level, a third classification level, and a fourth classification level.

[0033] Exemplarily, the classification level of vegetation is determined in the following manner:

[0034] Since corn, sunflower and alfalfa are distributed in a large area, the above plants are classified into the first classification level, and other plants are classified into other 1 of the first classification level. The first classification level corresponds to the layer level1 segmentation scale 210, and the classification level is set, for example, its level number is set to 3-5.

[0035] The first classification level "Other 1" is further subdivided to obtain the second classification level, corresponding to the layer level2 segmentation scale 180, the ground objects are Pinus sylvestris var. mongolica, Salix psammophila, bare soil and Other 2, and its level number is set to 2 to 4; in addition, since the ground object category of this level is a subset of "Other 1", Threshold--Edit Condition--Relation to objects--Super-Objects--Existence of Other 1(1) can be set to 1.

[0036] The second classification level "Other 2" is further subdivided to obtain the third classification level, corresponding to the layer level3 segmentation scale 70, the landforms are Artemisia ordosica and grassland, and its level number is set to 1 to 3; in addition, since the landform category of this level is a subset of "Other 2", it is necessary to set restriction conditions, and Threshold--Edit Condition--Relation to objects--Super-Objects--Existence of Other 2 (1) can be set to 1.

[0037] The classification results of the first three classification levels are merged to obtain the fourth classification level, which corresponds to the layer level 4 segmentation scale 70, and its level number is set to 0 to 2. In addition, since the feature category of this level is a subset of other classification levels, it is necessary to set restrictions for each feature category. For example, corn is the recognition result of level 1, and the distance between level 4 and level 1 is 3, so its restriction condition is Threshold--Edit Condition--Relation to objects--Super-Objects--Existence of corn (3) = 1; Pinus sylvestris is the recognition result of level 2, and the distance between level 4 and level 2 is 2, so its restriction condition is Existence of Pinus sylvestris (2) = 1; Artemisia ordosica is the recognition result of level 3, and the distance between level 4 and level 3 is 1, so its restriction condition is Existence of Artemisia ordosica (1) = 1.

[0038] The classification level is set in this step for multi-level classification.

[0039] S103: extracting feature information at each segmentation scale, and selecting vegetation features using random forests based on the extracted feature information, identifying the selected vegetation features, and obtaining identification results.

[0040] Optionally, the feature information includes spectral features, vegetation indexes and texture features, wherein the spectral features include the mean of each channel of RGB images and HSV images; the vegetation index includes the overgreening index EXG, the difference enhanced vegetation index DEVI, the difference vegetation index VDVI, the normalized green-red difference index NGRDI, the normalized green-blue difference index NGBDI and the red-green ratio index RGRI; the texture features include the mean, variance, contrast, difference, information entropy, correlation, homogeneity and second-order moment of each channel of the RGB image and the HSV image.

[0041] The vegetation index calculation formula is as follows:

[0042] EXG=(2G-RB) / (R+G+B)

[0043] DEVI=G / 3G+R / 3G+B / 3G

[0044] VDVI=(2G-RB) / (2G+R+B)

[0045] NGBDI=(GB) / (G+B)

[0046] NGRDI=(GR) / (G+R)

[0047] RGRI=R / G

[0048] Among them, R, G, and B are the grayscale values ​​of the red, green, and blue channel pixels of the drone's visible light image, respectively.

[0049] Optionally, for each segmentation scale, a random forest is used to perform feature importance analysis and select the optimal feature combination.

[0050] Exemplarily, random forest is used to perform feature importance ranking analysis. Through repeated experiments, when the number of features is 30, the classification accuracy is high and relatively stable. Therefore, the top 30 features are selected in each layer to sort the objects.

[0051] Optionally, a support vector machine is used to identify and extract the selected vegetation features to obtain an identification result.

[0052] Exemplarily, the optimal feature combination is selected for classification and recognition, and the classification includes two steps: classification training and classification implementation.

[0053] Classification training is applied to the first three classification levels. Before object recognition, supervised classification training is required. In the corresponding layer, supervised classification function is used for training. In operation, select train, and select the top 30 preferred features.

[0054] It should be noted that since the feature categories of level 2 and level 3 are subdivisions of the classification results of the previous layer, restrictions need to be set in Condition, namely Existence of Other 1 (1) = 1 and Existence of Other 2 (1) = 1.

[0055] Exemplarily, the classification is performed corresponding to three classification levels. At the corresponding layer, classification is performed using Apply in the supervisedclassification function.

[0056] It should be noted that since the feature categories of level 2 and level 3 are subdivisions of the classification results of the previous layer, restrictions also need to be set in Condition, namely Existence of other 1 (1) = 1 and Existence of other 2 (1) = 1.

[0057] S104: Merge the recognition results according to the classification level of vegetation to determine the spatial distribution of different vegetation.

[0058] Exemplarily, in level 4, the results of the first three classification levels are combined, and the classification results are analyzed to determine the spatial distribution of vegetation.

[0059] It should be noted that since the feature categories of level 2 and level 3 are subdivisions of the classification results of the previous layer, it is necessary to set restrictions in Condition, for example, Existence of Corn (3) = 1, Existence of Pinus sylvestris (2) = 1.

[0060] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A method for extracting vegetation information from visible light images, characterized in that: include: Acquire a visible light image of the vegetation test area; Use multi-scale segmentation algorithm to segment visible light images, determine the best segmentation scale for different vegetation, and determine the classification level of vegetation based on the best segmentation scale for different vegetation; Extracting feature information at each segmentation scale, and selecting vegetation features using random forests based on the extracted feature information, identifying the selected vegetation features, and obtaining identification results; wherein the feature information includes spectral features, vegetation indexes, and texture features; The identification results are merged according to the classification level of vegetation to determine the spatial distribution of different vegetation.

2. The method for extracting vegetation information from visible light images as claimed in claim 1, characterized in that: The step of identifying the selected vegetation features to obtain an identification result specifically includes: The support vector machine is used to identify and extract the selected vegetation features to obtain the recognition results.

3. The method for extracting vegetation information from visible light images as claimed in claim 1, characterized in that: The spectral characteristics specifically include: The mean of each channel of the RGB image and the HSV image; the HSV image is obtained by converting the visible light image to the HSV image.

4. The method for extracting vegetation information from visible light images as claimed in claim 1, characterized in that: The vegetation index specifically includes: Excess greenness index EXG, difference enhanced vegetation index DEVI, difference vegetation index VDVI, normalized green-red difference index NGRDI, normalized green-blue difference index NGBDI and red-green ratio index RGRI.

5. The method for extracting vegetation information from visible light images as claimed in claim 1, characterized in that: The texture features specifically include: The mean, variance, contrast, difference, information entropy, correlation, homogeneity and second-order moment of each channel of RGB images and HSV images.

Images