Investigation method of unmanned aerial vehicle for afforestation verification

Abstract

The invention discloses an investigation method of an unmanned aerial vehicle for afforestation verification, relates to the technical field of afforestation verification, and solves the technical problems that a large number of images shot by the unmanned aerial vehicle are directly utilized without being processed, follow-up analysis is interfered, meanwhile, the obtained images are not systematically analyzed, and verification errors are caused. According to the method, the image is enhanced through mean filtering denoising and histogram equalization, the image quality is effectively improved, interference of noise on subsequent feature extraction, matching and other links is reduced, feature points are extracted based on the constructed DOG image, the pre-selected image is scientifically screened by using the Euclidean distance, matching is further subdivided according to the number and distribution of the feature points, and the image quality is improved. On the basis of the spliced image, by means of GIS software, manual sketching or automatic classification can be flexibly adopted to measure the area of a region, an NDVI formula is utilized to accurately quantify the vegetation coverage, the vegetation growth coverage condition is judged according to a set interval, and the vegetation type is accurately recognized according to image features.

Description

Technical Field

[0001] The present invention relates to the technical field of afforestation verification, and in particular to a survey method for afforestation verification using an unmanned aerial vehicle. Background Art

[0002] With the development of drone technology, it has become a trend to use drones equipped with cameras to obtain images to assist in verification. In afforestation verification work, traditional verification methods mostly rely on manual field surveys, which is time-consuming, labor-intensive, inefficient, and difficult to achieve comprehensive, detailed and accurate verification of large afforestation areas. Therefore, drones are also used in afforestation verification work.

[0003] At present, drones often use visible light cameras to obtain high-resolution color images, and some combine multispectral cameras to obtain multi-band images to assist in analyzing information such as vegetation growth vitality. In the field of image processing, there are conventional operation technologies such as mean filtering for image denoising and histogram equalization for image enhancement. At the same time, in terms of image feature point extraction and matching, there are methods such as constructing DOG images to extract feature points and comparing feature point descriptor vectors based on the Euclidean distance formula to determine the degree of matching. Image stitching and subsequent analysis methods based on stitched images using GIS software, vegetation index calculation, and image feature identification of tree species are also gradually being applied to related fields. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a survey method for afforestation verification using unmanned aerial vehicles, which solves the problem that a large number of images taken by unmanned aerial vehicles are directly used without being processed, which will interfere with subsequent analysis, and that the obtained images are not systematically analyzed, which will cause errors in the verification.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a survey method for afforestation verification using a drone, the method specifically comprising the following steps:

[0006] Step 1: Obtain verification information and images taken by the drone;

[0007] Step 2: performing a preprocessing operation on the acquired captured image to obtain a preprocessed captured image;

[0008] Step 3: extracting the image feature points corresponding to the obtained pre-processed captured image, constructing a DOG image, selecting the feature points based on the similarity values ​​of the pixels of the pre-processed captured image, and screening the pre-processed captured image according to the feature points to obtain a pre-selected image;

[0009] Step 4, analyzing the obtained pre-selected image, segmenting the pre-selected image, and comprehensively selecting the matching image according to the number and distribution of feature points in the segmented pre-selected image, and splicing the matching images to obtain a spliced ​​image;

[0010] Step five: Based on the obtained stitched images, the verification area is investigated and analyzed, and verification information is generated by verifying the area area, vegetation coverage and vegetation type of the verification area.

[0011] As a further solution of the present invention, the verification information in step one includes the verification area and the drone route.

[0012] As a further solution of the present invention, the preprocessing operation in step 2 includes image denoising and image enhancement processing.

[0013] As a further solution of the present invention, the specific method of extracting image feature points in step 3 is:

[0014] All pre-processed images are obtained and labeled as i, where i=1, 2, ..., j, where j represents the number of pre-processed images. Then, any group of pre-processed images is obtained as a standard and recorded as the analysis object, and a DoG image of the analysis object is constructed. At the same time, the pixel points corresponding to the DoG image are obtained and labeled as n, where n=1, 2, ..., m, where m represents the number of pixel points. The pixel value of pixel point n is calculated and recorded as Hn. Then, the pixel values ​​Hn of the pixel points are sorted from small to large, and the pixel point with the smallest pixel value Hn is selected as the feature point and recorded as u.

[0015] Similarly, the feature points of all pre-processed captured images i are analyzed and selected as iu.

[0016] As a further solution of the present invention, the specific method of screening the pre-processed captured images to obtain the pre-selected images in step 3 is:

[0017] The pre-processed image corresponding to label i=1 is obtained as the image to be analyzed, and the corresponding feature point 1u is obtained. Then, the image to be analyzed is analyzed in the order of labels i=2, 3, ..., j, and the specific calculation method is:

[0018] Get the feature point 1u of the image to be analyzed, and get the descriptor vector of the feature point 1u, denoted as A=(a1, a2, ..., a o ), then obtain the feature point corresponding to the preprocessed image with label i=2 and record it as 2u, and similarly obtain the corresponding descriptor vector and record it as B=(b1, b2, ..., b o ), and according to the Euclidean distance formula The characteristic point distance d is calculated, and then the distance d is compared with a threshold value to screen the pre-processed captured image to obtain a pre-selected image, and the value of the threshold value is set by the operator.

[0019] As a further solution of the present invention, the specific method of comparing the distance with the threshold in step 3 is:

[0020] If the distance d is greater than the preset value, it means that the pre-processed image of the corresponding label does not match the image to be analyzed. On the contrary, if the distance d is less than the preset value, it means that the pre-processed image of the corresponding label matches the image to be analyzed. At the same time, the corresponding pre-processed image is obtained and recorded as the pre-selected image.

[0021] In this way, all the labeled pre-processed captured images are analyzed and pre-selected images are selected.

[0022] As a further solution of the present invention, the specific method of obtaining the spliced ​​image in step 4 is:

[0023] Obtain all pre-selected images and images to be analyzed, and divide the pre-selected images and the images to be analyzed into nine equal parts to obtain segmented images, then match the pre-selected segmented images with the segmented images to be analyzed in the segmentation order, obtain the number and distribution of feature points in the pre-selected segmented images, and simultaneously obtain the number and distribution of feature points in the segmented images to be analyzed. If the number of feature points in the pre-selected segmented images is the same as that in the segmented images to be analyzed, then mark the pre-selected segmented images as matching images, otherwise, if any of the groups are different, then do not select them;

[0024] In this way, matching images are obtained by analyzing according to the labels, and the obtained matching images are sorted in order to obtain a spliced ​​image.

[0025] As a further solution of the present invention, the specific method of generating the verification information in step 5 is:

[0026] Acquire the stitched image, measure the area of ​​the corresponding verification area in the stitched image based on GIS software, and generate area information;

[0027] Secondly, the spectral information of the image of the inspection area corresponding to the stitched image is used to evaluate the vegetation coverage by calculating the NDVI. At the same time, the obtained NDVI is judged to obtain the NDVI judgment interval. The judgment interval is specifically set by the operator based on historical data, and the NDVI of the current inspection area is matched with the judgment interval to generate vegetation coverage information.

[0028] Then, the tree species for afforestation are identified based on the color, texture, shape and other characteristics of the stitched image, and vegetation type information is generated.

[0029] Beneficial Effects

[0030] The present invention provides a survey method for afforestation verification using a drone. Compared with the prior art, it has the following beneficial effects:

[0031] The present invention enhances images through mean filtering denoising and histogram equalization, effectively improves image quality, reduces noise interference on subsequent feature extraction, matching and other links, enhances image details and contrast, and facilitates accurate analysis. It extracts feature points based on the construction of DOG images, uses Euclidean distance to scientifically screen pre-selected images, and further subdivides and matches according to the number and distribution of feature points, splices them in sequence to form a complete and practical spliced ​​image, highly restores the scene of the verification area, and lays a solid foundation for subsequent verification and analysis. Based on the spliced ​​image, with the help of GIS software, manual delineation or automatic classification can be flexibly used to accurately measure the area of ​​the area; the NDVI formula is used to accurately quantify vegetation coverage, and the vegetation growth and coverage situation is scientifically judged according to the set interval; the vegetation type is accurately identified according to the image characteristics, and the afforestation area is comprehensively and accurately checked from the spatial scale to the vegetation growth state and vegetation composition, which greatly improves the efficiency and quality of afforestation verification work and reduces manpower input and errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a diagram of the steps of the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.

[0034] For example, see Figure 1 The present application provides a method for investigating afforestation using a drone, which method specifically includes the following steps:

[0035] Step 1: Obtain verification information and images taken by the drone at the same time. The verification information includes the verification area and the drone route. The most common way to take drone images is to use a visible light camera, which can obtain high-resolution color images. In some cases, it will be combined with a multispectral camera. The multispectral camera can obtain images in multiple bands, such as red, green, blue, and near-infrared bands. For example, the reflectivity of vegetation in the near-infrared band is higher. By analyzing the images in this band, the growth vitality of vegetation can be evaluated.

[0036] In step 2, the acquired captured image is preprocessed to obtain a preprocessed captured image, and the preprocessing operation includes denoising and enhancing the captured image. Specifically, the image denoising is performed by a mean filtering method, and the image enhancement is performed by a histogram equalization method. The above operations are all existing technologies and will not be described in detail here.

[0037] Step three, extract the image feature points corresponding to the obtained preprocessed captured image, construct a DOG image, select feature points based on the similarity values ​​of the pixels of the preprocessed captured image, and filter the preprocessed captured image according to the feature points to obtain a preselected image.

[0038] Get all the preprocessed images and label them as i, and i = 1, 2, ..., j, where j represents the number of preprocessed images, and the label here is a random label. Then get any group of preprocessed images as the standard and record them as the analysis object, and construct the DoG image of the analysis object. The DoG image is obtained by subtracting Gaussian blurred images of different scales. Specifically, we assume that we have an image pyramid, which consists of multiple groups (octaves), each group contains multiple layers (layers). In a group, the standard deviation of the Gaussian kernel between adjacent layers is different. For example, the standard deviation of the Gaussian kernel of the first layer is σ, the second layer is kσ, and the third layer is k 2 σ, where k is the proportionality coefficient, usually taken as The DoG image is the result of subtracting two adjacent layers of Gaussian blurred images, that is, DoG(x, y, σ) = G(x, y, kσ) - G(x, y, σ), where G(x, y, σ) is a Gaussian blurred image at coordinates (x, y) with a standard deviation of σ. At the same time, the pixel points corresponding to the DoG image are obtained and labeled as n, and n = 1, 2, ..., m, where m represents the number of pixel points, and the pixel value of pixel point n is calculated and recorded as Hn. Then, the pixel values ​​Hn of the pixels are sorted from small to large, and the pixel point with the smallest pixel value Hn is selected as the feature point and recorded as u;

[0039] Similarly, the feature points of all pre-processed captured images i are analyzed and selected as iu, and the pre-processed captured image corresponding to the label i=1 is obtained as the image to be analyzed, and the corresponding feature point 1u is obtained, and then the image to be analyzed is calculated and analyzed in the order of the labels i=2, 3, ..., j, and the specific calculation method is:

[0040] Get the feature point 1u of the image to be analyzed, and get the descriptor vector of the feature point 1u, denoted as A=(a1, a2, ..., a o ), then obtain the feature point corresponding to the preprocessed image with label i=2 and record it as 2u, and similarly obtain the corresponding descriptor vector and record it as B=(b1, b2, ..., b o), and according to the Euclidean distance formula The feature point distance d is calculated, and then the distance d is compared with a threshold value, and the value of the threshold value is set by the operator;

[0041] If the distance d is greater than the preset value, it means that the pre-processed image of the corresponding label does not match the image to be analyzed. On the contrary, if the distance d is less than the preset value, it means that the pre-processed image of the corresponding label matches the image to be analyzed. At the same time, the corresponding pre-processed image is obtained and recorded as the pre-selected image. By analogy, all pre-processed images of the labels are analyzed and the pre-selected image is selected.

[0042] Step 4: Analyze the obtained pre-selected image, segment the pre-selected image, and obtain a matching image by comprehensive selection based on the number and distribution of feature points in the pre-selected image after segmentation, and simultaneously splice the matching images to obtain a spliced ​​image.

[0043] Acquire all pre-selected images and images to be analyzed, and the images to be analyzed and the pre-selected images are obtained by the same group analysis, specifically, by distance comparison, and the pre-selected images and the images to be analyzed are divided into nine equal parts to obtain segmented images, and the segmented images include the pre-selected segmented images and the segmented images to be analyzed, and then match the pre-selected segmented images with the segmented images to be analyzed according to the segmentation order, and the matching method is: obtain the number of feature points and the distribution of feature points in the pre-selected segmented images, and at the same time obtain the number of feature points and the distribution of feature points in the segmented images to be analyzed, if the number of feature points in the pre-selected segmented images is the same as that in the segmented images to be analyzed, then mark the pre-selected segmented images as matching images, otherwise if any group is different, then do not select them;

[0044] Similarly, the matching images are analyzed according to the labels, and the matching images are sorted in order to obtain a stitched image. For example, after the image to be analyzed with the label 1 is analyzed, the matching image obtained is the preprocessed image with the label 5, and then the image to be analyzed with the label 5 is analyzed, and the matching image of the image to be analyzed 5 is analyzed and selected, and so on and so forth to obtain the final stitched image.

[0045] Step five: Based on the obtained stitched images, the verification area is investigated and analyzed, and verification information is generated by verifying the area area, vegetation coverage and vegetation type of the verification area.

[0046] Get the stitched image, and measure the area of ​​the corresponding verification area in the stitched image based on GIS software, and generate area information. The specific method can be manual delineation or automatic classification. Manual delineation is to use the mouse to outline the boundary of the afforestation in the software based on the color, texture and other characteristics of the image, and the software will automatically calculate the area. Automatic classification uses algorithms such as machine learning to separate the afforestation area from the background based on the spectral characteristics of the image, and then calculate the area;

[0047] Secondly, the vegetation coverage is evaluated by calculating the vegetation index (such as the normalized difference vegetation index - NDVI) using the spectral information of the image of the verification area corresponding to the stitched image. The calculation formula of NDVI is (near infrared band reflectance - red light band reflectance) / (near infrared band reflectance + red light band reflectance). The value of NDVI is between -1 and 1. The higher the value, the higher the vegetation coverage. At the same time, the obtained NDVI is judged, and the specific judgment method is: obtain the NDVI judgment interval, and the judgment interval is specifically set by the operator according to historical data, and match the NDVI of the current verification area with the judgment interval, and generate vegetation coverage information at the same time;

[0048] Then, the afforestation tree species are identified based on the color, texture, shape and other characteristics of the stitched image, and vegetation type information is generated. Different tree species appear differently in the image. For example, the crown shape of broad-leaved trees such as poplars is relatively broad, while the crown shape of coniferous trees such as pine trees is relatively slender. At the same time, the leaf colors of different tree species also differ in the image. During the growing season, the leaves of broad-leaved trees are lighter green, while those of coniferous trees are darker green.

[0049] Embodiment 2, as Embodiment 2 of the present invention, is implemented on the basis of Embodiment 1, and differs from Embodiment 1 in the following aspects:

[0050] There are different ways to identify vegetation types. Specifically, tree species can be identified by combining the spectral characteristics of multispectral images. By establishing a tree species spectral feature library and using classification algorithms, the types and distribution of afforestation tree species can be determined.

[0051] Embodiment 3, as the embodiment 3 of the present invention, focuses on combining the implementation processes of embodiment 1 and embodiment 2 for implementation.

[0052] Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0053] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A survey method for afforestation verification using a drone, characterized in that: The method specifically comprises the following steps: Step 1: Obtain verification information and images taken by the drone; Step 2: performing a preprocessing operation on the acquired captured image to obtain a preprocessed captured image; Step 3, extracting the image feature points corresponding to the obtained pre-processed captured image, constructing a DOG image, selecting feature points based on the similarity values ​​of the pixels of the pre-processed captured image, and screening the pre-processed captured image according to the feature points to obtain a pre-selected image; Step 4, analyzing the obtained pre-selected image, segmenting the pre-selected image, and comprehensively selecting the matching image according to the number and distribution of feature points in the segmented pre-selected image, and splicing the matching images to obtain a spliced ​​image; Step five: Based on the obtained stitched images, the verification area is investigated and analyzed, and verification information is generated by verifying the area area, vegetation coverage and vegetation type of the verification area.

2. The method for investigating afforestation using a drone according to claim 1, characterized in that: The verification information in step one includes the verification area and the drone route.

3. The method for investigating afforestation using a drone according to claim 1, characterized in that: The preprocessing operation in step 2 includes image denoising and image enhancement processing.

4. The method for investigating afforestation using a drone according to claim 1, characterized in that: The specific method of extracting image feature points in step 3 is: All pre-processed images are obtained and labeled as i, where i=1, 2, ..., j, where j represents the number of pre-processed images. Then, any group of pre-processed images is obtained as a standard and recorded as the analysis object, and a DoG image of the analysis object is constructed. At the same time, the pixel points corresponding to the DoG image are obtained and labeled as n, where n=1, 2, ..., m, where m represents the number of pixel points. The pixel value of pixel point n is calculated and recorded as Hn. Then, the pixel values ​​Hn of the pixel points are sorted from small to large, and the pixel point with the smallest pixel value Hn is selected as the feature point and recorded as u. Similarly, the feature points of all pre-processed captured images i are analyzed and selected as iu.

5. The method for investigating afforestation using a drone according to claim 1, characterized in that: The specific method of screening the pre-processed captured images to obtain the pre-selected images in step 3 is: The pre-processed image corresponding to label i=1 is obtained as the image to be analyzed, and the corresponding feature point 1u is obtained. Then, the image to be analyzed is analyzed and calculated in the order of label i=2, 3, ..., j. The specific calculation method is: Get the feature point 1u of the image to be analyzed, and get the descriptor vector of the feature point 1u, denoted as A=(a1, a2, ..., a o ), then obtain the feature point corresponding to the preprocessed image with label i=2 and record it as 2u, and similarly obtain the corresponding descriptor vector and record it as B=(b1, b2, ..., b o ), and according to the Euclidean distance formula The characteristic point distance d is calculated, and then the distance d is compared with a threshold value to screen the pre-processed captured image to obtain a pre-selected image, and the value of the threshold value is set by the operator.

6. The method for investigating afforestation using a drone according to claim 5, characterized in that: The specific method of comparing the distance with the threshold in step 3 is: If the distance d is greater than the preset value, it means that the pre-processed image of the corresponding label does not match the image to be analyzed. On the contrary, if the distance d is less than the preset value, it means that the pre-processed image of the corresponding label matches the image to be analyzed. At the same time, the corresponding pre-processed image is obtained and recorded as the pre-selected image. In this way, all the labeled pre-processed captured images are analyzed and pre-selected images are selected.

7. The method for investigating afforestation using a drone according to claim 1, characterized in that: The specific method of obtaining the spliced ​​image in step 4 is: Obtain all pre-selected images and images to be analyzed, and divide the pre-selected images and the images to be analyzed into nine equal parts to obtain segmented images, then match the pre-selected segmented images with the segmented images to be analyzed in the segmentation order, obtain the number and distribution of feature points in the pre-selected segmented images, and simultaneously obtain the number and distribution of feature points in the segmented images to be analyzed. If the number of feature points in the pre-selected segmented images is the same as that in the segmented images to be analyzed, then mark the pre-selected segmented images as matching images, otherwise, if any of the groups are different, then do not select them; In this way, matching images are obtained by analyzing according to the labels, and the obtained matching images are sorted in order to obtain a spliced ​​image.

8. The method for investigating afforestation using a drone according to claim 1, characterized in that: The specific method of generating the verification information in step 5 is: Acquire the stitched image, measure the area of ​​the corresponding verification area in the stitched image based on GIS software, and generate area information; Secondly, the spectral information of the image of the inspection area corresponding to the stitched image is used to evaluate the vegetation coverage by calculating the NDVI. At the same time, the obtained NDVI is judged to obtain the NDVI judgment interval. The judgment interval is specifically set by the operator based on historical data, and the NDVI of the current inspection area is matched with the judgment interval to generate vegetation coverage information. Then, the tree species for afforestation are identified based on the color, texture, shape and other characteristics of the stitched image, and vegetation type information is generated.

Images