A method for extracting forest land information in power transmission and transformation engineering construction area by remote sensing

Abstract

This invention discloses a remote sensing method for extracting forest information in power transmission and transformation engineering construction areas, belonging to the field of forest information remote sensing extraction technology. The method includes selecting high-resolution satellite remote sensing images of the main autumn crops harvested in the task area based on the phenological characteristics of the task area; segmenting the images to generate image object layers; calculating the normalized vegetation index (NVI); determining three thresholds: NVI, brightness, and gray-level co-occurrence matrix contrast; and classifying the image objects according to these thresholds to obtain forest information. This invention uses high-resolution remote sensing images and extracts forest information based on forest characteristics. It has the advantages of simple principle, strong applicability, high accuracy, and streamlined technical process, and can be applied to large-scale forest information extraction.

Description

Technical Field

[0001] This invention relates to the field of remote sensing extraction technology for forest land information, and more specifically, to a method for remote sensing extraction of forest land information in power transmission and transformation engineering construction areas. Background Technology

[0002] Power transmission and transformation projects involve numerous construction sites along long distances, complex construction cycles, diverse topography, and significant terrain variations, often traversing ecologically sensitive and vulnerable areas. Timely and accurate extraction of forest land information and monitoring of forest land changes are crucial for ecological environmental protection management and maintaining the stability of forest ecosystems during the construction phase of power transmission and transformation projects. Traditional methods for acquiring forest land information are inefficient, lack timeliness, and are costly. Remote sensing information technology, with its real-time and rapid advantages, has become an important means of acquiring forest land information. Examples include medium spatial resolution satellite remote sensing imagery such as GF-1, MODIS, Sentinel-2, and Landsat series, as well as high spatial resolution satellite remote sensing imagery such as IKONOS, QuickBird, and WorldView series. Medium-resolution imagery data has low spatial resolution and is not effective for identifying scattered forest land; high-resolution satellite remote sensing imagery has significant advantages in forest land information extraction. In recent years, researchers have conducted research on forest land information extraction methods based on different types of remote sensing imagery data and various approaches. However, these studies are mainly based on medium- and high-resolution imagery, with limited research on forest land extraction methods for high-resolution, especially sub-meter resolution, remote sensing imagery. There is a lack of technical methods that can be applied on a large scale in operational settings. In remote sensing imagery, forest land is easily confused with crop and grassland information. Therefore, effectively distinguishing forest land from crops and grassland using remote sensing imagery is the key and challenge in extracting forest land distribution information. Summary of the Invention

[0003] Technical problems to be solved

[0004] To address the problems existing in the prior art, the purpose of this invention is to provide a remote sensing method for extracting forest information in power transmission and transformation engineering construction areas. This method can utilize high-resolution (better than 1 meter) remote sensing images to quickly and accurately extract forest information in operational applications, thus making it more suitable for practical use.

[0005] Technical solution

[0006] To solve the above problems, the present invention adopts the following technical solution.

[0007] The forest information referred to in this invention includes the spatial location, extent, and area information of the forest land.

[0008] A remote sensing method for extracting forest land information in power transmission and transformation project construction areas includes the following steps:

[0009] Step 1: Acquire remote sensing images based on the phenological conditions of the task area.

[0010] High-resolution satellite remote sensing imagery of the mission area was acquired. The imagery time was selected based on the phenological characteristics of the mission area, specifically during the autumn harvest of major crops. At this time, the woodlands had not yet begun to turn yellow, while the grasslands and farmland had already begun to wither, minimizing interference from grasslands and farmland in the extraction of woodland information.

[0011] The image includes blue, green, red, and near-infrared bands.

[0012] The images should have undergone preprocessing such as radiometric correction, band fusion, and geometric correction.

[0013] Step 2: Image Segmentation

[0014] The image of the task area is first segmented into a chessboard pattern, and then multi-scale segmentation is performed based on the chessboard segmentation. An image object layer is then generated.

[0015] Step 3: Calculation of Normalized Difference Vegetation Index (NDVI)

[0016] Normalized Difference Vegetation Index (NDVI) data was obtained by calculating the image data.

[0017] The formula for calculating the Normalized Difference Vegetation Index (NDVI) is as follows:

[0018]

[0019] Wherein, NDVI is the normalized vegetation index, ρ NIR ρ represents the pixel surface reflectance value in the near-infrared band. R This represents the pixel surface reflectance value in the red light band;

[0020] The normalized vegetation index (NDI) of an image object is the average of the NDI values ​​of each pixel within the image object.

[0021] Step 4: Determine the threshold

[0022] Threshold Sample Selection: Select forest land image objects from the image data as sample plots for threshold selection. When selecting sample plots, it is required to include all different forest land types in the area as much as possible, including a certain number of forest lands within mountain shadows.

[0023] Thresholding objects: The thresholds in this method include the Normalized Difference Vegetation Index (NDVI) threshold, the HIS-intensity threshold, and the Gray-Level Co-occurrence Matrix Contrast (GLCM) threshold.

[0024] Normalized Difference Vegetation Index (NDVI) Threshold: The NDVI threshold is the average NDVI of all pixels within the image object. Two thresholds need to be selected, namely N1 and N2. N1 is the minimum NDVI value among the non-shaded forest samples, and N2 is the maximum NDVI value among the shaded forest samples.

[0025] Brightness (HIS-intensity) threshold: I1, where I1 is the maximum value of the HSI transformation-brightness value of the forest sample.

[0026] Gray co-occurrence matrix contrast (GLCM) threshold: C1 is half of the sum of the maximum and minimum values ​​in the forest sample.

[0027] Step 5: Threshold Classification

[0028] Forest land extraction is performed on image objects based on thresholds.

[0029] (1) Extract image objects with a normalized vegetation index (NDVI) value greater than or equal to N1 and a brightness (HIS-intensity) value less than or equal to I1 as forest land;

[0030] (2) Extract image objects with a normalized vegetation index (NDVI) value greater than or equal to N2, a brightness (HIS-intensity) value less than or equal to I1, and a gray-level co-occurrence matrix contrast (GLCM) value less than or equal to C1 as forest land.

[0031] The forest land information extracted in steps (1) and (2) is the forest land information extracted by this method.

[0032] Vegetation has a high NDVI value, while forest land, due to the height and unevenness of the trees, often forms certain shadow areas. Therefore, forest land objects have lower brightness values ​​compared to cultivated land and grassland objects. Step (1) above is to extract forest land by setting a reasonable threshold based on the NDVI value and brightness characteristics of forest land. (2) is because forest land in the mountain shadow has a low NDVI value, and extraction according to the method in (1) will result in some forest land in the mountain shadow being missed. Since the gray-level co-occurrence matrix contrast (GLCM contrast) value of forest land objects is lower than that of cultivated land and grassland, a second extraction is performed based on the GLCM contrast value of the image objects. This ensures the completeness of forest land extraction.

[0033] Beneficial effects

[0034] Compared with the prior art, the advantages of this invention are:

[0035] Based on the NDVI, brightness, and texture characteristics of forest land, this invention can quickly and accurately extract forest land information. It has the advantages of simple principle, streamlined technology, and ease of business application.

[0036] This invention selects images of the main autumn crop harvest period based on the phenological characteristics of the task area, which can greatly eliminate the interference of cultivated land and grassland on forest land information and improve the accuracy of forest land information extraction.

[0037] This invention can be widely applied in northern my country, and has a broad scope of application. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the process of the present invention;

[0039] Figure 2 This is a schematic diagram of the mission area image of the present invention;

[0040] Figure 3 This is a schematic diagram of the image extraction results of the present invention. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0042] Example 1:

[0043] Taking the construction area of ​​the "Chenguantun-Xiaoyingzi 220kV Transmission and Transformation Project" in Lulong County, Qinhuangdao City, Hebei Province as an example, combined with Figure 1 The present invention will be further described in detail with reference to specific embodiments, but this is not intended to limit the present invention.

[0044] A remote sensing method for extracting forest land information in power transmission and transformation project construction areas includes the following steps:

[0045] Step 1: Obtain remote sensing images based on the phenological conditions of the task area.

[0046] High-resolution satellite remote sensing images of the main crops in the mission area during the autumn harvest season were selected based on their phenological characteristics. At this time, the main crops have already begun to wither and turn yellow, while the forest land has not yet begun to turn yellow, which can greatly reduce the interference of crops on forest land information extraction.

[0047] In late September, corn, the main crop in Lulong County, Qinhuangdao City, Hebei Province, began to be harvested. High-resolution satellite remote sensing images obtained from late September to early October were used as the base images for forest land information extraction. In this embodiment, a 0.3-meter resolution remote sensing image from Beijing-3 on September 25, 2022, was selected as the base image for forest land information extraction.

[0048] The imagery includes blue, green, red, and near-infrared bands. The imagery has undergone preprocessing including radiometric correction, band fusion, and geometric correction. The mission area imagery is as follows: Figure 2 .

[0049] Step 2, image segmentation.

[0050] The remote sensing image is segmented to generate an image object layer.

[0051] Image segmentation is performed in two steps. The first step is chessboard segmentation, with the segmentation object size controlled between 3 and 5 meters. In this example, the image resolution is 0.3 meters, and the segmentation object size is set to 10 pixels. The second step is multi-scale segmentation based on the chessboard segmentation in the first step. In this example, the multi-scale segmentation scale is 200, the shape parameter is 0.5, and the compactness parameter is 0.5.

[0052] Step 3: Calculate the Normalized Difference Vegetation Index (NDVI).

[0053] The image data is numerically calculated to obtain the Normalized Difference Vegetation Index (NDV) raster data. The NDC for an image object is the average NDC value of each pixel within the image object.

[0054] Step 4: Determine the threshold.

[0055] Determine the thresholds needed for the next classification step.

[0056] Thresholds include three categories: Normalized Difference Vegetation Index (NDVI) threshold, HIS-intensity threshold, and Gray-Level Co-occurrence Matrix (GLCM) contrast threshold. HIS-intensity and GLCM contrast are image pair features. HIS-intensity is the brightness value of the RGB bands of the image after HIS transformation; here, it is the average brightness value of all pixels within the image object. The Gray-Level Co-occurrence Matrix (GLCM) is a method for describing texture by studying the spatial correlation characteristics of gray levels. GLCM contrast directly reflects the contrast between the brightness of a pixel value and its neighboring pixels. In this example, the GLCM contrast is the contrast at 0°.

[0057] Forest samples are selected from the image object layer. The number of forest samples should be no less than 20, and the samples should include all forest types in the area. In this embodiment, 20 samples are selected.

[0058] The Normalized Difference Vegetation Index (NDVI) threshold has two values, N1 and N2. N1 is the minimum NDVI value for the selected non-shaded forest land samples. N2 is the minimum NDVI value for the selected shaded forest land samples. In this example, N1 is 0.56 and N2 is 0.4.

[0059] The brightness (HIS-intensity) threshold I1 is the minimum value of the brightness statistics of the forest sample. In this embodiment, the value of I1 is 0.16.

[0060] The gray-level co-occurrence matrix contrast (GLCM contrast) threshold C1 is half the sum of the maximum and minimum values ​​of the statistical values ​​of the gray-level co-occurrence matrix contrast of the forest samples. In this embodiment, the value of C1 is 500.

[0061] Step 5, Threshold Classification

[0062] Threshold classification is performed on image objects based on the selected threshold. Threshold classification is performed in two steps.

[0063] Image objects with a Normalized Difference Vegetation Index (NDVI) value greater than or equal to N1 and a Brightness Index (HIS-intensity) value less than or equal to I1 are extracted as woodland.

[0064] Image objects with a Normalized Difference Vegetation Index (NDVI) value greater than or equal to N2, a HIS-intensity value less than or equal to I1, and a Gray-Level Co-occurrence Matrix Contrast (GLCM) contrast less than or equal to C1 were extracted as woodland. The extraction results are as follows: Figure 3 .

[0065] Step six, accuracy verification.

[0066] In order to effectively verify and evaluate the method of the present invention, the accuracy of the forest land information extracted in the examples was verified.

[0067] Accuracy was verified by comparing the forest land information extracted using manual visual interpretation. Forest land was extracted manually from the task area, and the extracted forest land vectors were spatially superimposed with those extracted using this method. The area of ​​forest land extracted manually is denoted as MJ_R, and the area extracted using this method is denoted as MJ_M. The overlapping area is denoted as MJ_R&M.

[0068] Calculate the producer accuracy and user accuracy separately.

[0069] The formula for calculating producer accuracy is:

[0070]

[0071] The formula for calculating user accuracy is:

[0072]

[0073] In this embodiment, the producer accuracy is 93.5% and the user accuracy is 93%, which meets the needs of practical applications.

[0074] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concepts, should be covered within the scope of protection of the present invention.

Claims

1. A method for remote sensing extraction of forest land information in power transmission and transformation engineering construction areas, characterized in that, Includes the following steps: Step 1: Acquire remote sensing images based on the phenological conditions of the mission area. Acquire high-resolution satellite remote sensing images of the mission area. The imaging time of the satellite images is selected according to the phenological characteristics of the mission area, specifically during the autumn harvest of major crops. The remote sensing images include blue band, green band, red band, and near-infrared band. The acquired remote sensing images are preprocessed by radiometric correction, band fusion, and geometric correction. Step 2: Image segmentation. First, the image of the task area is segmented into a chessboard pattern. Then, multi-scale segmentation is performed on the chessboard pattern to generate the image object layer. Step 3: Normalized Difference Vegetation Index (NDVI) calculation. The NDVI is calculated from the image data to obtain the normalized vegetation index (NDI). The formula for calculating the NDVI is: Wherein, NDVI is the normalized vegetation index, ρNIR is the pixel surface reflectance value in the near-infrared band, and ρR is the pixel surface reflectance value in the red band; the normalized vegetation index of the image object is the average value of the normalized vegetation index of each pixel in the image object. Step 4: Determine the thresholds. Select forest image objects from the image data as sample plots for threshold selection. The determined thresholds include the Normalized Difference Vegetation Index (NDVI) threshold, brightness threshold, and gray-level co-occurrence matrix (GLCM) contrast threshold; where: The normalized vegetation index threshold is set to two thresholds, N1 and N2, where N1 is the minimum NDVI value of the non-shaded forest sample and N2 is the maximum NDVI value of the shaded forest sample. The brightness threshold is I1, where I1 is the maximum value among the HSI transform brightness values ​​of the forest sample; The contrast threshold of the gray-level co-occurrence matrix is ​​C1, where C1 is half of the sum of the maximum and minimum values ​​in the forest sample; Step 5: Threshold classification. Based on thresholds, forest land is extracted from the image objects. This includes the following steps: S1: Extract image objects with a normalized vegetation index value greater than or equal to N1 and a brightness value less than or equal to I1 as woodland; S2: Extract image objects with a normalized vegetation index value greater than or equal to N2, a brightness value less than or equal to I1, and a gray-level co-occurrence matrix contrast value less than or equal to C1 as forest land; The forest land information extracted in step S1 and step S2 together constitute the final extracted forest land information; Step Six: Accuracy Verification. Accuracy verification is performed by comparing the forest land information extracted through manual visual interpretation. Specifically, the forest land in the task area is extracted through manual visual interpretation. The forest land vector extracted by manual visual interpretation is spatially superimposed with the forest land vector extracted by this method. The forest land area extracted by manual visual interpretation is denoted as MJ_R, and the forest land area extracted by this method is denoted as MJ_M. The overlapping area is denoted as MJ_R&M. Producer accuracy and user accuracy are calculated separately.

2. The method for remote sensing extraction of forest land information in power transmission and transformation engineering construction areas according to claim 1, characterized in that, The high-resolution satellite remote sensing imagery acquired in step one is a sub-meter level remote sensing imagery with a spatial resolution better than 1 meter.

3. The method for remote sensing extraction of forest land information in power transmission and transformation engineering construction areas according to claim 1, characterized in that, In step two, the size of the chessboard segmentation object is controlled at 3-5 meters, the segmentation scale of multi-scale segmentation is 200, the shape parameter is 0.5, and the compactness parameter is 0.

5.

4. The method for remote sensing extraction of forest land information in power transmission and transformation engineering construction areas according to claim 1, characterized in that, In step four, the number of selected forest land image sample plots shall not be less than 20, and the sample plots shall cover all forest land types within the task area, including forest land samples within mountain shadows.

5. The method for remote sensing extraction of forest land information in power transmission and transformation engineering construction areas according to claim 1, characterized in that, The gray-level co-occurrence matrix contrast in step four is the contrast value in the 0° direction.

6. The method for remote sensing extraction of forest land information in power transmission and transformation engineering construction areas according to claim 1, characterized in that, In step six, the formula for calculating producer accuracy is: The formula for calculating user accuracy is: .

Images