Vegetation identification method and device based on unmanned aerial vehicle image, readable medium and program

Abstract

The invention relates to a vegetation identification method and device based on an unmanned aerial vehicle image, a readable medium and a program, and the method comprises the steps: obtaining the unmanned aerial vehicle image of a to-be-identified region, and calculating a constructed RGBSI vegetation index based on the brightness values or reflectivity of the red light wave band, the green light wave band and the blue light wave band of the image, and extracting a vegetation region from the image of which the RGBSI vegetation index is calculated by using a threshold segmentation method. The RGBSI vegetation index constructed by the method combines the strong reflection of green vegetation in a green light wave band and the absorption characteristics of the green vegetation in red and blue wave bands, when the method is used for vegetation identification, an image histogram of a vegetation index grey-scale map image calculated by RGBSI has a good double-peak form under the condition of medium and high vegetation coverage, a threshold can be rapidly determined by using a maximum entropy threshold method, and the method can be applied to vegetation identification. And the threshold value is generally close to 0. The overall extraction precision of identifying the vegetation-containing image by using the RGBSI index is more than 90%, the extraction precision is high, and the extraction effect of vegetation with medium and high vegetation coverage is better.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and particularly relates to a vegetation recognition method, device and program based on unmanned aerial vehicle (UAV) images. Background Art

[0002] In recent years, UAV technology has developed rapidly. Compared with traditional remote sensing technologies such as aerospace, the advantages of UAV remote sensing are: (1) It belongs to low-altitude remote sensing and is rarely interfered by atmospheric factors; (2) It has low cost, simple and convenient operation, and is flexible and mobile; (3) The images can be transmitted in real time and have high spatial resolution. Therefore, UAVs make it possible to quickly and efficiently obtain vegetation information in small areas.

[0003] At present, there are already a variety of vegetation indices for vegetation recognition. However, different vegetation indices usually have different recognition accuracies for different types of images. At present, vegetation indices based on the red, green, and blue light bands are relatively lacking. There is an urgent need for a vegetation index applicable to different vegetation coverages to improve the accuracy of small-scale vegetation information extraction. Summary of the Invention

[0004] The first object of the present invention is to provide a vegetation recognition method based on UAV images in view of the deficiencies of the above-mentioned existing technologies.

[0005] To achieve the above technical object, the present invention adopts the following technical solutions:

[0006] A vegetation recognition method based on UAV images, comprising:

[0007] Obtaining UAV images of the area to be recognized, and calculating the RGBSI vegetation index based on the brightness values or reflectances of the red, green, and blue light bands of the images:

[0008]

[0009] In the formula, R, G, and B are respectively the brightness values or reflectances of the red, green, and blue light bands in the images;

[0010] Extracting the vegetation area from the images for calculating the RGBSI vegetation index by using the threshold segmentation method.

[0011] As a preferred implementation manner, the UAV images are digital orthophotos.

[0012] As a preferred implementation manner, calculating the RGBSI vegetation index based on the reflectances of the red, green, and blue light bands of the images.

[0013] As a preferred implementation manner, using the maximum entropy method or extracting the vegetation area from the images.

[0014] As a preferred embodiment, the vegetation coverage degree in the UAV images is ≥ 30%.

[0015] As a preferred embodiment, the vegetation coverage degree in the UAV images is ≥ 70%.

[0016] The second object of the present invention is to provide a computer device, including a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0017] The second object of the present invention is to provide a computer-readable storage medium, on which a computer program / instructions are stored, and when the computer program / instructions are executed by a processor, the steps of the above method are implemented.

[0018] The third object of the present invention is to provide a computer program product, which includes a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the above method are implemented.

[0019] The present invention analyzes the spectral characteristics of vegetation and non-vegetation. Combining the characteristics that green vegetation has strong reflection in the green light band and strong absorption in the blue and red light bands, the strong reflection effect of vegetation in the green light band is enhanced by multiplying the green light band in visible light by two and then performing square conversion; among non-vegetation categories, the green light band has the characteristic that the difference from the blue and red light bands is small, and is smaller than the difference of the vegetation category. The square of twice the green light band minus the square of the sum of the red and blue light bands can be used to further expand the recognition range of vegetation types in visible light images, narrow the recognition range of non-vegetation types, weaken the overlapping part between vegetation information and non-vegetation information, and reduce the influence of the overlap on the discrimination of vegetation information, thereby constructing a new visible light vegetation index RGBSI (Green-Red-Blue Square Index).

[0020] The RGBSI vegetation index constructed by the present invention combines the strong reflection of green vegetation in the green light band and its absorption characteristics in the red and blue light bands. When used for vegetation recognition, the vegetation index grayscale image calculated by RGBSI has a good bimodal shape in the histogram under medium and high vegetation coverage conditions. The maximum entropy threshold method can be used to quickly determine the threshold, and the threshold is generally near 0. The overall extraction accuracy of using the RGBSI index to recognize vegetation-containing images is above 90%, with high extraction accuracy, and better extraction effect for vegetation with medium and high vegetation coverage. Description of the Drawings

[0021] Figure 1 is the digital orthophoto image of the study area with different vegetation coverage degrees.

[0022] Figure 2 is the threshold of each vegetation index in the study area.

[0023] Figure 3 It is a schematic diagram of extracting vegetation using different vegetation indices in different study areas.

[0024] Figure 4 It is the evaluation result of vegetation extraction accuracy. Specific implementation manners

[0025] The technical solution of the present invention will be further elaborated below in conjunction with the accompanying drawing description and specific implementation manners.

[0026] Example 1

[0027] 1. Study area and image acquisition

[0028] The study area selected in this example is located in Tongshan District, Xuzhou City. As Figure 1 shown, the image acquisition time is November 2023. The UAV model is Huace Hornet BB4 quadcopter UAV, the flight altitude is 200m, the ground resolution is 2.5cm, and the number of pixels in a single image is 8184 5640, and more than 100 digital orthophoto maps (DOM) are obtained.

[0029] In this example, three relatively typical areas with different vegetation coverage degrees are selected as the study areas, and the orthophoto images are obtained as Figure 1 shown. Study area 1 is an area with low vegetation coverage, and the vegetation coverage degree is less than 30%. The main ground object types are roads, buildings, forests, and bare soil; Study area 2 is an area with medium vegetation coverage, and the vegetation coverage degree is between 30% and 70%. The main ground object types are roads, buildings, forests, bare soil, and grasslands; Study area 3 is an area with high vegetation coverage, and the vegetation coverage degree is greater than 70%. The main ground object types are roads, forests, and bare soil.

[0030] 2. Vegetation index calculation

[0031] Based on the constructed vegetation index RGBSI (green - red - blue square index) in the visible light band, the RGBSI index of each pixel of each image is calculated. The formula of the RGBSI index is as follows:

[0032]

[0033] In the formula, R, G, and B respectively represent the pixel brightness values or reflectances of the red, green, and blue bands in the UAV visible light image; in this example, reflectance is used for calculation.

[0034] To compare the reliability and applicability of RGBSI for vegetation information extraction, this embodiment introduces the relatively typical visible light vegetation indices that have been proposed for accuracy comparison, including 5 visible light vegetation indices: Visible-band Difference Vegetation Index - VDVI, Excess Green - EXG, Excess Red - EXR, Excess Green minus excess Red index - EXGR, and Color Index of Vegetation Extraction - CIVE (see Table 1).

[0035] Table 1 5 visible light vegetation indices

[0036]

[0037] Note: R: Red channel; G: Green band; B: Blue channel, r: Normalized result of the red channel; g: Normalized result of the green channel; b: Normalized result of the blue channel. Among them: , , and the ranges of r, g, and b are [0, 1].

[0038] 3. Determination of vegetation index thresholds

[0039] The gray-scale distribution map can only relatively simply distinguish the vegetation and non-vegetation areas of 6 vegetation indices. In this embodiment, the maximum entropy method is used to determine the thresholds of each vegetation index in 3 study areas. The results are as Figure 2 shown. In study area 1, the thresholds of RGBSI, VDVI, and EXG are all near 0, the thresholds of EXR and EXGR are near 0.1, and the threshold of GIVE is the highest, about 18.76; in study area 2, the thresholds of RGBSI, VDVI, and EXG are all near 0, the thresholds of EXR and EXGR are near 0.1, and the threshold of GIVE is the highest, about 18.77; in study area 3, RGBSI, VDVI, EXG, and EXR are in the range of 0.05 - 0.1, the threshold of EXGR is about 0.01, and the threshold of GIVE is about 18.73. Generally speaking, in the three areas, the thresholds of RGBSI, VDVI, and EXG are near 0, the thresholds of EXR and EXGR are near 0.1, and the threshold of GIVE is near 18.76.

[0040] 4. Vegetation information extraction

[0041] The images of the three study areas were segmented using the calculated thresholds to extract vegetation information, and the results are as follows Figure 3 shown (a - f is the extraction result of study area 1, g - l is the extraction result of study area 2, m - r is the extraction result of study area 3).

[0042] In study area 1, GIVE and EXGR failed to identify the small - scale vegetation under the right - hand building, while RGBSI, VDVI, EXG, and EXR could better identify the small - scale vegetation under the right - hand building. In study area 2, EXR misidentified the two houses on the right as vegetation, while the other five indices identified the two houses on the right as non - vegetation, and EXR also misidentified the darker - green vegetation as non - vegetation, while the other five vegetation indices did not have this situation. EXR, GIVE, and EXGR misidentified the relatively sparse vegetation as non - vegetation, while RGBSI, VDVI, and EXG could correctly identify the relatively sparse vegetation. In study area 3, EXG, EXR, EXGR, and GIVE missed the identification of the densely vegetated part in the lower - right corner, while RGBSI and VDVI did not have this situation.

[0043] 5. Accuracy evaluation

[0044] To evaluate the accuracy of vegetation extraction based on six vegetation indices, 300 points were randomly selected from each of the three images. Using a human - computer interaction method, each pixel of the images was divided into vegetation and non - vegetation areas, and the accuracy evaluation of the vegetation information extraction results was carried out. The vegetation accuracy, non - vegetation accuracy, overall accuracy, and Kappa coefficient were calculated. The accuracy evaluation results of the three areas are shown in Figure 4 . The overall accuracies of the three vegetation indices RGBSI, VDVI, and EXG were all above 90%, and the Kappa coefficients were also above 0.8, indicating that these three visible - light vegetation indices are better at extracting vegetation and non - vegetation information. The overall accuracies of the three vegetation indices EXR, EXGR, and CIVE were between 80% and 90%, and the Kappa coefficients were between 0.6 and 0.8, indicating that these three vegetation indices are relatively worse than the above - mentioned three in extracting vegetation and non - vegetation information.

[0045] For the accuracy evaluation of RGBSI, with the increase in vegetation coverage, the vegetation correct rate increased from 91.67% to 92.78% and then to 94.74%; the total correct rate increased from 90.67% to 92.67% and then to 94.00%; the Kappa coefficient also increased from 0.8077 to 0.8481 and then to 0.8713. Generally speaking, the average vegetation correct rate and kappa coefficient of RGBSI are the highest among the six vegetation indices, indicating that RGBSI has a better effect on extracting vegetation information. And RGBSI has a better effect on extracting vegetation with medium and high vegetation coverage.

Claims

1. A vegetation recognition method based on drone images, characterized in that, Including: Obtain the UAV image of the area to be recognized, and calculate the RGBSI vegetation index based on the brightness values or reflectivities of the red, green, and blue bands of the image: ； In the formula, R, G, and B are the brightness values or reflectivities of the red, green, and blue bands in the image respectively; Use the threshold segmentation method to extract the vegetation area from the image for calculating the RGBSI vegetation index.

2. The method according to claim 1, characterized in that, The UAV image is a digital orthophoto image.

3. The method according to claim 1, wherein Calculate the RGBSI vegetation index based on the reflectivities of the red, green, and blue bands of the image.

4. The method according to claim 1, characterized in that, Use the maximum entropy method to extract the vegetation area from the image.

5. The method according to claim 1, wherein The vegetation coverage in the UAV image is ≥ 30%.

6. The method according to claim 1, characterized in that, The vegetation coverage in the UAV image is ≥ 70%.

7. A computer device, comprising a memory, a processor, and a computer program stored on the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having computer programs / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images