Method for extracting secondary forest window from remote sensing image based on supervised classification

Abstract

The invention discloses a supervised classification-based method for extracting a secondary forest window from a remote sensing image, and relates to the technical field of forest window remote sensing identification, and the method comprises the steps: firstly collecting and preprocessing a high-precision remote sensing image, then setting a random sample belt, recording coordinates, importing the image, and then interpreting the image through a supervised classification tool, a test algorithm is constructed to continuously adjust a classification system, finally post-processing is performed on a classification result to complete forest window identification and vectorization output, the secondary forest window is accurately and efficiently extracted from a remote sensing image, important data support is provided for research and monitoring of secondary forests, and the structure and dynamic change of a secondary forest ecosystem can be known.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing identification of forest gaps, and particularly to a method for extracting secondary forest gaps from remote sensing images based on supervised classification. Background Art

[0002] In recent years, secondary forests formed by the degradation of primary forests due to disturbances have become the main body of China's forest resources. Inducing the positive succession of degraded forests and restoring their inherent ecological service functions are the main goals of national and Hebei ecological security construction and forestry development. Promoting near-natural forestry management that mimics and conforms to nature has become the main means of secondary forest management. As the most common small-scale disturbance in the forest regeneration process, forest gaps have become the main object in the process of near-natural forestry management. Clearly defining the characteristics of forest gaps and mastering the distribution characteristics of forest gap patterns are the premise and foundation for the near-natural management of secondary forests using forest gaps.

[0003] Traditional plot-scale forest gap investigation methods are restricted by high labor costs and small coverage areas, and it is difficult to reflect and obtain the distribution characteristics of forest gap patterns. In recent years, the development of remote sensing interpretation technology has provided technical support for the study of the distribution characteristics of forest gap patterns. However, at present, unsupervised classification methods are used for forest gap extraction from remote sensing images, and the accuracy of the obtained forest gap interpretation results is limited, and there is inevitably a lack in terms of scientificity. Especially in natural secondary forests with diverse tree species compositions and complex age structures, the large number of categories obtained by unsupervised classification leads to the usually unusable final forest gap interpretation results. Therefore, a method that can accurately extract forest gaps from secondary forest remote sensing images is needed. Summary of the Invention

[0004] To solve the above problems, the present invention provides a method for extracting secondary forest gaps from remote sensing images based on supervised classification, aiming to obtain a forest gap identification method with high interpretation result accuracy, low cost, and wide coverage.

[0005] To achieve the above object, the technical solution adopted by the present invention is as follows: A method for extracting secondary forest gaps from remote sensing images based on supervised classification. First, a high-precision remote sensing image is pre-constructed, random sample belts are set to record coordinates and the image is imported. Then, the supervised classification algorithm is used to interpret the image and a verification algorithm is constructed to continuously adjust the classification system. Finally, post-processing is performed on the classification results to complete forest gap identification and vectorization output, so as to accurately and efficiently extract secondary forest gaps from the remote sensing image and form a global data support that can analyze the structure and dynamic changes of the secondary forest ecosystem.

[0006] Further preferably, it specifically includes the following steps: a. Collection and preprocessing of remote sensing images: Collect high-precision remote sensing images that can be used for secondary forest gap extraction, and perform preprocessing such as radiometric calibration, geometric correction, and atmospheric correction; b. Set up random sample belts: Conduct on-site surveys of secondary forests, use high-precision GPS to record the coordinates of on-site forest gaps and the central points of forest patches composed of different tree species, and import these coordinates into the remote sensing image. c. Supervised classification and interpretation: Use the supervised classification tool of ERDAS software, take various patches surveyed on-site on the image as training samples for the supervised classification standard, establish a classification system, complete the comprehensive interpretation of the image, construct a verification algorithm to verify the classification results, and continuously adjust the classification system until the verification coefficient of the classification results is greater than 0.8. d. Post-processing of classification results: Conduct clustering and removal analysis on the output classification results to remove the "salt and pepper phenomenon", remove patches that are too small and too large according to the forest canopy gap range standard and the height of the secondary forest canopy, complete the forest gap identification work, and perform vectorization output.

[0007] Furthermore, the high-precision remote sensing image should meet the following conditions: (1) The spatial resolution is higher than 2 meters; (2) The acquisition time is between June and September; (3) The cloud cover of the image is less than 5%.

[0008] Furthermore, the positioning error of the high-precision GPS should be less than 1 meter.

[0009] Furthermore, record the number of on-site forest gaps and forest patches composed of different tree species, with no less than 20 patches for each type.

[0010] Furthermore, the forest canopy gap range standard is that D / H is between 0.23 and 3.23, where D is the diameter of the forest gap and H is the height of the secondary forest canopy.

[0011] Furthermore, in step a, the geometric correction of the remote sensing image is carried out using 10 to 40 control points in the 1:5000 topographic map, and the root mean square error of the correction is less than one pixel.

[0012] Furthermore, the specific process of constructing a verification algorithm to verify the classification results in step c is as follows: a. Construct a confusion matrix and count the corresponding relationship between the classification results and the true reference data; b. Calculate the overall classification accuracy; c. Calculate the expected agreement; d. Calculate the verification coefficient: Subtract the expected agreement from the overall classification accuracy to get the difference f, and then divide it by the difference g obtained by subtracting the expected agreement from 1; e. Ensure that the verification coefficient is greater than 0.8 by continuously adjusting the classification system.

[0013] Further, the calculation process of the overall classification accuracy is as follows: the number of correctly classified pixels is within the total number of pixels.

[0014] Further, the calculation process of the expected consistency is as follows: a. Assume there are N categories. For the i-th category in sequence, count the number of samples of this category in the true data, denoted as n t,i , and the number of samples of this category in the classification result, denoted as n r,i ; b. For each category, calculate the product of its true sample number and the sample number in the classification result, n t,i ×n r,i ; c. Sum up the products of all categories to obtain S; d. Determine the total number of pixels in the dataset, denoted as M, and calculate its square value M 2 ; e. Divide the summation result S by the square of the total number of pixels M 2 to obtain the expected consistency.

[0015] Compared with the prior art, the technical progress achieved by the present invention is as follows: through supervised classification, the present invention combines forest gap remote sensing and field investigations, solves the problem of misclassification of forest gaps caused by different subjective experiences of researchers, greatly improves the accuracy of forest gap extraction in secondary forest images, and can be applied to the large-scale forest gap identification work of high-resolution remote sensing images of secondary forests. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention.

[0017] In the drawings: Figure 1 is a schematic flow chart of the method for extracting secondary forest forest gaps from remote sensing images based on supervised classification of the present invention; Figure 2 is a diagram of the operation process of the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] 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 with reference to the drawings. Embodiment 1

[0019] Secondary forests formed by the degradation of primary forests have become the main body of forest resources in my country. Taking the secondary forests in Northeast China as an example, the original top-level community broad-leaved Korean pine forests have been transformed into natural secondary forests composed of various maple species, Juglans mandshurica, maple birch, Tilia amurensis, Quercus mongolica, etc. after complex interference. The diverse tree species composition further increases the difficulty of identifying forest windows in remote sensing images, limiting the near-natural management of secondary forests with forest window regeneration as the core. In order to test the effectiveness of the present invention, a typical secondary forest located in the Dahu work area of ​​Dasuhe State Forest Farm in Dasuhe Township, the southern end of Qingyuan County, Liaoning Province, with a total area of ​​1,350 hectares, was selected for forest window extraction. Example 2

[0020] A method for extracting secondary forest gaps from remote sensing images based on supervised classification, comprising the following steps: a. Remote sensing image collection and preprocessing: Select the Gaofen-2 satellite image of the study area, with a spatial distribution rate of 1 meter for panchromatic and 4 meters for multispectral. The acquisition time is July 12, 2023, and the cloud cover is 1%, which meets the image standard suitable for secondary forest window extraction in step 1. Use ENVI software to perform image radiometric calibration, atmospheric correction and other preprocessing. Select 20-30 control points from the 1:5000 topographic map to perform geometric correction of the image. The root mean square error (RMSE) of the correction is less than one pixel. Use wavelet transform to fuse the multispectral image with its corresponding panchromatic image to improve the overall resolution of the image.

[0021] b. Set up random sampling strips: Conduct field surveys of secondary forests, set up three survey plots in the study area, each 5 kilometers long, record the coordinates of various forest types and forest window center points along the way, and obtain 30 field patch locations of each type.

[0022] c. Supervised classification and interpretation: ERDAS supervised classification module is used to classify the images. According to the location of each type of field patch, the classification standard of forest window, Mongolian oak forest, broad-leaved mixed forest, poplar forest, and larch forest patches is trained. Then the overall image classification is completed according to the standard, and the verification algorithm is constructed to verify the classification results. The specific process is as follows: a. Construct a confusion matrix to count the correspondence between the classification results and the real reference data; b. Calculate the overall classification accuracy: the number of correctly classified pixels is the total number of pixels; c. Calculate expected consistency: a. Assume that there are N categories, and count the number of samples of the i-th category in the real data n t,i , and the number of samples of this category in the classification result, recorded as n r,i ; b. For each category, calculate the product of the number of true samples and the number of classified result samples n t,i×n r,i ; c. Sum the products of all categories to obtain S; d. Determine the total number of pixels in the dataset, denoted as M, and calculate its square value M 2 ; e. Divide the summation result S by the square of the total number of pixels M 2 to obtain the expected consistency; d. Calculate the test coefficient: Subtract the expected consistency from the overall classification accuracy to obtain the difference f, and then divide it by the difference g obtained by subtracting the expected consistency from 1; e. Ensure that the test coefficient is greater than 0.8 by continuously adjusting the classification system.

[0023] Verify the consistency of the classification results of the test method for the same research object, continuously adjust the classification results to improve the test coefficient, and finally the test coefficient of the classification results is 0.87.

[0024] d. Post - processing of classification results: Use the spatial analysis tools in Arcgis software to perform clustering and removal analysis on the classification results, optimize the classification results, and reduce the "salt - and - pepper phenomenon" in the image. Based on the average tree height of the secondary forest canopy in the study area (20m) and the upper and lower limits of the forest gap size (0.23 ≤ D / H ≤ 3.23), determine the forest gap area range as "16 - 3257m 2 ", and screen the forest gap patches in the classification results accordingly, and finally perform vectorization output.

[0025] Finally, it should be noted that the above - mentioned are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A method for extracting secondary forest gaps from remote sensing images based on supervised classification, characterized in that: First, high-precision remote sensing images are pre-constructed, and random sample strips are set to record coordinates and import images. Then, supervised classification algorithms are used to interpret images, and verification algorithms are constructed to continuously adjust the classification system. Finally, the classification results are post-processed to complete forest gap identification and vectorized output, so as to accurately and efficiently extract secondary forest gaps from remote sensing images and form global data support that can analyze the structure and dynamic changes of secondary forest ecosystems.

2. The method for extracting secondary forest gaps from remote sensing images based on supervised classification according to claim 1, characterized in that: The specific steps include: a. Remote sensing image collection and preprocessing: collect high-precision remote sensing images that can be used for secondary forest gap extraction, and perform preprocessing such as radiation calibration, geometric correction, and atmospheric correction; b. Set up random sampling strips: Conduct field surveys of secondary forests, use high-precision GPS to record the coordinates of forest windows and the center points of forest patches composed of different tree species, and import these coordinates into remote sensing images; c. Supervised classification and interpretation: Using the supervised classification tool of ERDAS software, various patches of field survey on the image are used as training samples of the supervised classification standard, a classification system is established, a comprehensive interpretation of the image is completed, a test algorithm is constructed to test the classification results, and the classification system is continuously adjusted until the test coefficient of the classification result is greater than 0.8; d. Post-processing of classification results: cluster and remove the "salt and pepper phenomenon" on the output classification results. According to the canopy and gap range standards and the secondary canopy height, remove patches that are too small and too large, complete the forest gap identification work, and perform vectorized output.

3. The method for extracting secondary forest gaps from remote sensing images based on supervised classification according to claim 2, characterized in that: The high-precision remote sensing images should meet the following conditions: (1) Spatial resolution higher than 2 meters; (2) The collection period was between June and September; (3) The cloud cover in the image is less than 5%.

4. The method for extracting secondary forest gaps from remote sensing images based on supervised classification according to claim 2, characterized in that: The positioning error of the high-precision GPS should be less than 1 meter.

5. The method for extracting secondary forest gaps from remote sensing images based on supervised classification according to claim 2, characterized in that: The number of forest gaps and forest patches composed of different tree species was recorded, with no less than 20 patches of each type.

6. The method for extracting secondary forest gaps from remote sensing images based on supervised classification according to claim 2, characterized in that: The canopy gap range standard is D / H between 0.23 and 3.23, where D is the gap diameter and H is the secondary canopy height.

7. The method for extracting secondary forest gaps from remote sensing images based on supervised classification according to claim 2, characterized in that: In step a, the geometric correction of the remote sensing image is performed using 10 to 40 control points in a 1:5000 topographic map, and the root mean square error of the correction is less than one pixel.

8. The method for extracting secondary forest gaps from remote sensing images based on supervised classification according to claim 2, characterized in that: In step c, the specific process of constructing a test algorithm to test the classification results is as follows: a. Construct a confusion matrix to count the correspondence between the classification results and the real reference data; b. Calculate the overall classification accuracy; c. Calculate expected consistency; d. Calculate the test coefficient: subtract the expected consistency from the overall classification accuracy to get the difference f, and then divide it by 1 minus the expected consistency to get the difference g; e. Ensure that the inspection coefficient is greater than 0.8 by continuously adjusting the classification system.

9. The method for extracting secondary forest gaps from remote sensing images based on supervised classification according to claim 8, characterized in that: The overall classification accuracy calculation process is: the number of correctly classified pixels is equal to the total number of pixels.

10. The method for extracting secondary forest gaps from remote sensing images based on supervised classification according to claim 8, characterized in that: The calculation process of the expected consistency is: a. Assume that there are N categories, and count the number of samples of the i-th category in the real data n t,i , and the number of samples of this category in the classification result, recorded as n r,i ; b. For each category, calculate the product of the number of true samples and the number of classified result samples n t,i ×n r,i ; c. Sum the products of all categories to get S; d. Determine the total number of pixels in the data set, record it as M, and calculate its square value M 2 ; e. Divide the sum S by the square of the total number of pixels M 2 , and obtain the expected consistency.

Images