Hilly Terrain Feature Line Extraction Method, Hilly Terrain DEM Refined Production Method

Abstract

The present invention discloses a method for extracting hilly terrain feature lines and a method for refined production of hilly terrain DEM, which are applied to the technical field of feature line extraction. The spatial consistency and overlap degree between manually collected valley lines and automatically extracted results are compared; the time required for different map sheets by the automatic extraction method and the traditional manual collection method is statistically compared; in the planar and three-dimensional environments, the filtering effect of the valley area after optimizing and editing using the extracted feature lines is checked; quality assessment is carried out using checkpoints to detect whether the elevation values of the DEM result data on the valley lines meet the mathematical accuracy of the technical regulations.

Description

Technical Field

[0001] The present invention relates to the field of feature line extraction, and particularly to a method for extracting feature lines of hilly terrain and a method for fine production of hilly DEM. Background Art

[0002] The acquisition of global geographic information resources mainly uses optical stereo images, and terrain features are automatically extracted through dense matching of stereo image pairs. The production of DEM through the Digital Surface Model (DSM) can achieve the rapid acquisition of three-dimensional geographic information in large areas, and at the same time can make up for data missing or low resolution in areas with poor accessibility such as overseas or uninhabited areas. The high-resolution domestic satellite Ziyuan-3 (ZY-3) with stereo observation capabilities, with the advantages of high spatial resolution, high geometric positioning accuracy, and rich spectral bands, has now been used as the main data source for the global geographic information resource construction project and is widely used in the production of DEM in key global regions.

[0003] In the production of high-precision DEM, the filtering and editing of landforms such as hills, mountains, and high mountains must be assisted by terrain feature lines. Therefore, the accurate and rapid extraction of feature lines is an important prerequisite for ensuring the quality and efficiency of DEM results. Terrain feature lines include ridge lines and valley lines. As the dividing lines of terrain undulation changes, they describe the basic landform patterns of the land surface and play an important role in research such as geomorphological generalization, hydrological models, and watershed analysis, and play an extremely important role in the production of high-precision DEM. At present, in the process of production practice, the extraction of feature lines mainly has two methods: automatic extraction algorithms using terrain geometric features and manual stereo collection. In a stereo environment, although the method of relying on manual collection has high extraction accuracy, it requires a large amount of manpower and material resources and a long collection time. Although the automatic extraction algorithm based on terrain geometric features has a high degree of automation, factors such as valley vegetation coverage and occlusion cause incomplete or incorrect extraction of feature lines, which is likely to cause DEM editing errors. Therefore, when using DSM data for high-precision DEM production, existing research has not proposed a method for quickly and accurately extracting feature lines for high-resolution DSM data of hilly areas.

[0004] Therefore, how to provide a method for extracting feature lines of hilly terrain with high precision and a method for fine production of hilly DEM is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a method for extracting hilly terrain feature lines and a method for refined production of hilly terrain DEMs, comparing the spatial consistency and overlap degree between manually collected valley lines and automatically extracted results; statistically comparing the time required for different map sheets for the automatic extraction method and the traditional manual collection method; checking the filtering effect of the valley area after optimizing and editing using the extracted feature lines in the planar and three-dimensional environments; using checkpoints for quality assessment to detect whether the elevation values of the DEM result data on the valley lines meet the mathematical accuracy of the technical regulations.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for extracting hilly terrain feature lines, the specific steps include:

[0008] Construct a feature set and a sample library, and extract the vegetation coverage range at the valleys;

[0009] Use distance transformation and curvature analysis methods to obtain a preliminary extraction result of the valley feature lines;

[0010] Use topological analysis methods to optimize the continuity of the feature lines, and at the same time use local spatial relationships and empirical knowledge to remove heterogeneous line segments to obtain optimized valley feature lines.

[0011] Optionally, in the above method for extracting hilly terrain feature lines, the feature set includes texture features, elevation features, and spectral features, where the elevation features and spectral features are respectively obtained through preprocessing based on the DSM and DOM result data; the sample library stores all vegetation samples in the vegetation area to be extracted.

[0012] Optionally, in the above method for extracting hilly terrain feature lines, the specific steps for extracting the vegetation coverage range at the valleys are as follows:

[0013] Obtain the feature set required for remote sensing image classification, including spectral features, texture features, and elevation features, and perform band stacking to obtain a stacked image;

[0014] Perform unsupervised segmentation on the stacked image to generate segmentation objects;

[0015] Obtain the sample data required for remote sensing image classification;

[0016] Use the feature set and samples as the input for supervised classification to perform object-oriented image classification, and then obtain the vegetation surface distribution map of the target area, and further extract the vegetation coverage range at the valleys.

[0017] Optionally, in the above method for extracting hilly terrain feature lines, the steps for extracting valley feature lines are as follows:

[0018] Taking the object - oriented image classification result as the input, using the Euclidean distance, calculate the minimum distance value from all pixels within the valley vegetation area to the range boundary to obtain the distance field data;

[0019] Using the curvature analysis method to take the derivative of the distance field data, calculate the gradient data, and then perform re - classification and masking operations on the gradient data to extract the gradient maximum value area and obtain the preliminary extraction result of the valley feature line.

[0020] Optionally, in the above - mentioned method for extracting the hilly terrain feature line, the specific calculation steps of the distance field data are as follows:

[0021] Using the Euclidean distance to calculate the minimum distance value from all pixels within the vegetation coverage area at the valley in the binary classification map to the range boundary to obtain the distance field data;

[0022] De(x, y) = [(x2 - x1) 2 +(y2 - y1) 2 1 / 2 ;

[0023] In the formula, x i is the longitude coordinate value, and y i is the latitude coordinate value.

[0024] Optionally, in the above - mentioned method for extracting the hilly terrain feature line, the specific steps of the curvature analysis method are as follows:

[0025] Calculate the gradient magnitude S of the distance field data, and connect the points with larger gradient values to obtain the skeleton line;

[0026] S(x, y, De) = [(d de / d x ) 2 +(d de / d y ) 2 1 / 2 ;

[0027] In the formula, d de is the derivative of the distance field data, dx is the derivative in the longitude direction, and dy is the derivative in the latitude direction.

[0028] Optionally, in the above - mentioned method for extracting the hilly terrain feature line, for the continuity optimization of the valley feature line, introduce the grid expansion method to fill in the missing values and alleviate the problem of multiple small closed - loop skeleton lines; furthermore, introduce the topological refinement method to iteratively strip the outermost pixels to extract the valley line with a single - pixel width; perform buffer analysis on the valley line with a single - pixel width; then introduce the strategy of hanging nodes to remove the non - feature lines, and thus obtain the optimized valley feature line.

[0029] ​​A refined production method for hilly DEM using a method for extracting hilly terrain feature lines, the specific steps including:

[0030] Filter and edit the DSM tiled result data, reduce the surface elevation to the ground elevation, so as to obtain the preliminary filtered result of DEM;

[0031] Perform optimization processing on the preliminary filtered result. For the valley feature line, measure the vegetation height using forest gaps and assign elevation attributes to the feature line, and perform secondary filtering and editing processing on the DEM data through the three-dimensional feature line to obtain the DEM result data with reasonable expression of geomorphic form features.

[0032] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method for extracting hilly terrain feature lines and a refined production method for hilly DEM. Utilizing the correlation between surface cover features and geomorphic form features, an automatic extraction method for feature lines suitable for hilly terrain is proposed to replace the traditional manual stereoscopic acquisition operation mode. Through production tests on eight 1:50,000 standard tiled data of different types in different regions, it is proved from both qualitative and quantitative perspectives that the geometric position accuracy of the valley lines obtained by applying the method proposed in the present invention is high, the connectivity is good, the DEM result data after optimizing and editing using the feature lines fits well with the stereoscopic model, the geomorphic form is reasonably expressed, the data quality is good, and it fully meets the mathematical accuracy requirements. Starting from the actual production requirements, the present invention uses the technical means of remote sensing and GIS to propose a new operation method for key production links of DEM, improving production efficiency and reducing manual workload while ensuring product accuracy. This method can provide reference for high-precision DEM production in other key regions around the world. Description of the Drawings

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0034] Figure 1 It is a schematic diagram of the geographical distribution and topographic and geomorphic features of the test data in the embodiment of the present invention;

[0035] Figure 2 It is a schematic diagram of the description of the test data in the embodiment of the present invention;

[0036] Figure 3 It is a production flow chart of hilly DEM in the present invention;

[0037] Figure 4It is the flowchart of the ArcGIS python workflow of the present invention;

[0038] Figure 5 It is the schematic diagram of accuracy evaluation of the present invention;

[0039] Figure 6 It is the schematic diagram of comparison between the manually collected valley lines and the extraction results of the present invention;

[0040] Figure 7 It is the schematic diagram of comparison between the manual collection and the automatic extraction timeliness of the present invention;

[0041] Figure 8 It is the filtering effect diagram of valley vegetation in hilly areas of the present invention;

[0042] Figure 9 It is the result diagram of DEM achievement quality assessment of the present invention. Specific implementation manners

[0043] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0044] The embodiments of the present invention disclose a method for extracting hilly terrain feature lines and a method for fine production of hilly DEM. By using the correlation between surface cover features and geomorphic form features, a method for automatically extracting feature lines suitable for hilly terrain is proposed to replace the traditional manual stereo collection operation mode. Through production tests on eight 1:50,000 standard map-sheet data of different regions and different types, it is proved from both qualitative and quantitative perspectives that the geometric position accuracy of the valley lines obtained by applying the method proposed in the present invention is high, the connectivity is good, the DEM achievement data optimized and edited by using the feature lines is well fitted with the stereo model, the geomorphic form expression is reasonable, the data quality is good, and it fully meets the mathematical accuracy requirements. Starting from the actual production needs, the present invention proposes a new operation method for key production links of DEM by means of remote sensing and GIS technologies. On the premise of ensuring the product accuracy, the production efficiency is improved and the manual workload is reduced. This method can provide reference for the production of high-precision DEM in other key regions of the world.

[0045] In this embodiment, the hilly area of the Central African Republic (hereinafter referred to as "CAR") is selected as the test area. The CAR is a landlocked country in the central part of the African continent, with a land area of about 623,000 square kilometers, ranging between 14°25′W - 27°28′W and 2°13′N - 11°01′N ( Figure 1)。The altitude in the survey area is about 500 - 900 meters. Approximately 80% of the terrain is low mountains and hills, with relatively small undulations, indistinct ridge features, and relatively shallow valleys that gently transition to the slopes on both sides. The survey area belongs to the tropical savannah climate, with an average annual temperature of about 26°C. The rainy season is from May to October, and the dry season is from November to April of the following year. The hilly surface is mainly covered with sparse jungles and seasonal river swamps, and dense shrubs and arbors are distributed in the valleys.

[0046] It should be noted that the planar and elevation accuracies of ZY-3 stereo mapping satellite images are equivalent to or even better than those of similar international satellites, fully meeting the mapping accuracy requirements of 1:50,000. It has now been used as the main data source for the global geographic information resource construction project.

[0047] The data used in this embodiment includes ZY-3 stereo image pairs and DSM and DOM result data in the global geographic information resource construction project. The coordinate system is WGS-84, and the storage unit is the 1:50,000 standard quadrangle.

[0048] This embodiment adopts the methods of automatic matching and manual interactive editing to obtain DSM result data that fits the stereo model, has a reasonable geomorphic hillshading state, and meets the digital accuracy requirements, and uses it as the initial data for DEM production and the input features for object-oriented classification.

[0049] The DOM result data is generated by preprocessing the original panchromatic and multispectral images through orthorectification, fusion, histogram equalization, and cropping, etc., and is used for visual interpretation reference and the input features for object-oriented classification.

[0050] Considering the complexity and diversity of the hilly terrain in the survey area, in order to fully verify the feasibility of the method proposed in this study, eight 1:50,000 standard quadrangle data are selected, with uniform location distribution, covering different types of topographic and geomorphic features ( Figure 1 ). According to the feature characteristics of ground objects in the valley area and non-valley area, the hilly landforms in Central Africa are roughly divided into four categories. The vegetation cover types in the valley area include dense arbors and relatively sparse shrubs, and the ground object elements in the non-valley area include sparse grasslands and relatively sparse shrubs. See specifically Figure 2 .

[0051] It should be noted that considering the similarity and particularity of the surface cover characteristics of the hilly terrain in Central Africa, in the remote sensing image, the vegetation in the valley area grows densely and is mainly arbors, showing a zonal distribution characteristic. The vegetation on the slopes is sparse and mainly herbaceous plants. There are obvious differences in spectral, texture, and surface elevation between the valley area and the non-valley area. Using the remote sensing interpretation method, the valley range can be quickly and accurately extracted. Moreover, the automatic extraction of valley lines can be realized by using GIS spatial relationship analysis. Based on the above ideas, this study designed an automatic extraction process for feature lines of the hilly terrain in the Central Africa survey area, as Figure 3As shown in the figure, it mainly includes three steps: object - oriented classification, feature line extraction, and DEM secondary filtering and editing based on feature lines, and applies them to the DEM production process.

[0052] The specific process is as follows:

[0053] 1) Image segmentation and object - oriented classification. According to the DSM and DOM result data, construct a feature set (including spectral features, texture features, and elevation features) and a sample library, and extract the vegetation coverage range in the valley.

[0054] Furthermore, the specific steps to construct the feature set are as follows:

[0055] Band extraction: Extract three bands from the multi - spectral four - band data (including red band, blue band, green band, and near - infrared band), and recombine them into a new multi - spectral data (including red band, blue band, and green band).

[0056] Resampling: Use the mean clustering method to resample the data with a 2 - meter grid spacing to a 10 - meter grid spacing.

[0057] Texture feature calculation: Based on the multi - spectral data, use the gray - level co - occurrence matrix to calculate and generate texture features.

[0058] Mosaic: For adjacent original satellite images, use the seamless mosaic technology to generate mosaic lines, and use the mosaic lines to complete the digital surface model data mosaic.

[0059] Cropping: According to the provisions of GBT 13989 - 2012 and the requirements in the global geographic information resource construction project, the cropping range is extended 50 grid spacings outward from the minimum circumscribed rectangle of the four - corner coordinates of the inner map frame of the standard 1:50,000 sub - map.

[0060] Pixel bit conversion: Convert the pixel value range of the image from 32 - bit to 16 - bit depth.

[0061] Furthermore, use the remote - sensing image classification method to extract the vegetation coverage range (vegetation surface) in the valley. The main steps are as follows:

[0062] Step S1: Obtain the feature set required for remote - sensing image classification, including spectral features (derived from the digital orthophoto map result data, abbreviated as DOM result data, including 3 bands), texture features (calculated based on the digital orthophoto map using the gray - level co - occurrence matrix, including 1 band), and surface elevation information (derived from the digital surface model result data, abbreviated as DSM result data, including 1 band). The bands are stacked into an image with 5 bands.

[0063] Step S2: Perform unsupervised segmentation on the above - mentioned image to generate segmentation objects.

[0064] Step S3: Obtain the sample data required for remote sensing image classification.

[0065] Step S4: Use the feature set and samples as the input of supervised classification for object-oriented image classification, so as to obtain the vegetation area distribution map of the target area, and then extract the vegetation area range in the valley.

[0066] 2) Valley feature line extraction. Use the distance transformation and curvature analysis methods to obtain the preliminary extraction result of the valley line;

[0067] 3) Optimize the preliminary extraction result of the feature line. Use topological analysis and other methods to ensure the continuity of the feature line, and use local spatial relationships and empirical knowledge to remove heterogeneous line segments;

[0068] 4) Filter and edit the DSM mosaicking result data, reduce the surface elevation to the ground elevation to obtain the preliminary DEM filtering result; perform secondary filtering and editing on the DEM based on the feature line. Assign elevation information to the two-dimensional feature line, and use the three-dimensional feature line to optimize the preliminary vegetation height reduction result to reasonably represent the valley landform;

[0069] 5) Accuracy evaluation of the geometric position of the feature line extraction result and the DEM filtering effect.

[0070] In order to further optimize the above technical solution, the basic processing unit of object-oriented classification is the object generated by the segmentation algorithm. Compared with the traditional classification with pixels as the smallest processing unit, object-oriented classification is more suitable for the recognition of ground objects in high-spatial-resolution images. Through image segmentation, the entire image is segmented into independent regions or objects, and pixels with similar spectral and spatial attributes are merged into one object, and the difference between adjacent objects increases, so as to effectively remove noise interference and improve the classification accuracy. Object-oriented classification mainly includes two steps: segmentation and classification. Use the software in the existing technology, and the object-oriented classification module in the software provides segmentation and classification algorithms for high-spatial-resolution remote sensing images.

[0071] By setting the weight value of the segmentation feature, apply the segmentation process to the ZY-3 DOM and DSM result data. The segmentation scale size of the image is adjusted according to the width of the vegetation strip distribution. Generally speaking, the larger the width, the larger the segmentation scale parameter value. The feature set includes spectral, texture, and surface elevation features. In this embodiment, different combination schemes of key parameters are tested and the extraction effect of the valley area of each map sheet is analyzed to determine each parameter value. In addition, the embodiment of the present invention constructs a sample library through an interactive interface, and the sample library can be iteratively modified and re-predicted according to the prediction results.

[0072] To further optimize the above technical solution, the distance field data is regarded as terrain elevation data, and the feature line formed by connecting the pixels with the maximum slope of elevation change is the skeleton line. The Euclidean distance is used to calculate the minimum distance value from all pixels within the valley range in the binary classification map to the range boundary, obtaining the distance field data De, as shown in Equation (1).

[0073] De(x, y) = [(x2 - x1) 2 + (y2 - y1) 2 1 / 2 ; (1)

[0074] In the formula, x i is the longitude coordinate value, and y i is the latitude coordinate value.

[0075] The surface curvature analysis method is adopted to calculate the gradient magnitude S of the distance field data, and the points with larger gradient values are connected to obtain the skeleton line, as shown in Equation (2).

[0076] S(x, y, De) = [(d de / d x ) 2 + (d de / d y ) 2 1 / 2 ; (2)

[0077] In the formula, d de is the derivative of the distance field data, dx is the derivative in the longitude direction, and dy is the derivative in the latitude direction.

[0078] The preliminary extraction of the valley line is achieved through the above method. However, the preliminarily extracted valley line does not consider the actual surface geometry, and the extraction results mainly have problems of low spatial connectivity and misjudgment of feature lines. Therefore, it is essential to optimize and edit it. The low spatial connectivity is mainly manifested in two situations: disconnection between adjacent feature lines and connection of multiple closed-loop skeleton lines. The former is because in the result of surface curvature analysis, there are areas with relatively small gradients on the skeleton line, making it difficult to correctly divide them by threshold judgment. Therefore, on the premise that the distance between the endpoint coordinates of adjacent line segments is short, this embodiment introduces a grid expansion analysis strategy to connect adjacent line segments. For the latter situation of multiple closed-loop line segments, a topological refinement method is introduced to iteratively strip the outermost pixels, thereby extracting the valley line composed of a single pixel width. Due to the complex contour of the mountainous area, there are some short branches in the skeleton line proposed by the method based on distance transformation. However, according to the mathematical definition of the geomorphic form, these branches are not valley lines. This embodiment removes non-feature line segments by traversing all line segment endpoints and judging whether there are hanging nodes on the line segments, obtaining the vector valley line.

[0079] ​​Since the extracted valley range affects the accuracy of valley line extraction, in order to improve the accuracy of valley line extraction, the extraction results can be interactively post-processed based on empirical knowledge. In this study, ArcGIS Python is used to create a repeatable workflow as Figure 4 shown to meet the requirements of automatic extraction of feature lines. By comparing the data processing speed and effect of each module in the workflow through multiple experiments, the valley line extraction time is further shortened.

[0080] To further optimize the above technical solution, the DSM mosaicked data is filtered and edited to reduce the surface elevations of buildings, bridges, vegetation, etc. to the ground elevation, thereby obtaining DEM data. When producing DEM in hilly areas, it is also necessary to correctly represent the geomorphic morphological characteristics of valleys on the premise of ensuring that the mathematical accuracy meets the technical regulations. Since there are dense shrubs and arbors in the valleys and adjacent slopes in the hilly areas of Central Africa, obvious noises such as burrs and jump points will appear in the DEM data after conventional filtering and editing, and the valley morphological characteristics cannot be correctly represented. Therefore, it is necessary to optimize the results of preliminary filtering, that is, collect the feature lines reflecting the important valley morphology, measure the vegetation height using the forest gap and assign elevation attributes to the feature lines, and perform secondary filtering and editing on the DEM data through the three-dimensional feature lines to correct the influence of noises and obtain the DEM result data with reasonable expression of geomorphic morphological characteristics.

[0081] It should be noted that: the manual stereoscopic collection method is a feature line extraction method widely used in current production, which is a generally accepted method with relatively high accuracy. In this embodiment, the valley line collected manually is defined as the reference feature line, and the accuracy evaluation method based on the confusion matrix is used to analyze the geometric position relationship between the automatically extracted valley line and the reference feature line. The main idea of this method is to calculate the number of pixels where the extracted valley line and the reference feature line fall inside and outside their respective buffer zones, and construct a reference index according to each statistical result ( Figure 5 ). The specific method is as follows.

[0082] First, taking the reference feature line (manual collection result) as the matching standard, establish a buffer zone with a radius of 1 pixel or 2 pixels, and judge the spatial relationship between the automatically extracted valley line and the buffer zone. The confusion matrix is shown in Table 1. If the automatically extracted result falls into the buffer zone, it is considered to match the reference feature line, and the pixels falling into the buffer zone are recorded as TP1 (true positive), otherwise it is considered to not match the reference feature line, and the pixels outside the buffer zone are recorded as FP (false positive); second, taking the automatically extracted valley line as the matching standard, establish a buffer zone, and record the reference feature line pixels falling into the buffer zone as TP2 (true positive), otherwise it is considered to not match the extracted valley line, and is recorded as FN (false negative).

[0083] Table 1 Confusion Matrix

[0084]

[0085] Based on the analysis of the confusion matrix, two parameters, precision (Correct) and recall (Complete), can be defined. Among them, recall reflects the integrity of the automatically extracted valley lines, and precision reflects the accuracy of the automatically extracted valley lines. By combining these two parameters, the extraction quality (Quality) of the valley lines can be comprehensively evaluated, as shown in Equations (3)-(5) respectively.

[0086]

[0087]

[0088]

[0089] A variety of test methods are used to analyze and evaluate the extraction accuracy of valley lines and their application effects in the refined production of DEM, specifically including: 1) Comparing the spatial consistency and overlap degree between the manually collected valley lines and the automatically extracted results; 2) Statistically comparing the time required for the automatic extraction method and the traditional manual collection method for different map sheets; 3) Checking the filtering effect of the valley area after optimizing and editing using the extracted feature lines in the planar and three-dimensional environments; 4) Using checkpoints for quality assessment to detect whether the elevation values of the DEM result data on the valley lines meet the mathematical accuracy of the technical regulations.

[0090] According to the evaluation method of valley line extraction accuracy, the position accuracy of the automatic valley line extraction method proposed in this embodiment is quantitatively evaluated. The results are shown in Table 2 (Pixels within the buffer are recorded as matches, and pixels outside the buffer are recorded as non-matches. The measurement unit is pixel PX, and the pixel resolution is 10 meters). In this study, buffers with radii of 1 pixel and 2 pixels are generated for each piece of test data, and their accuracy results are analyzed.

[0091] Table 2 Accuracy Analysis of Valley Line Extraction Results

[0092]

[0093] Analysis and comparison of Table 2 show that: 1) When the buffer radius is 1 pixel, that is, the geometric positioning accuracy of the automatically extracted valley line is 1 pixel, the extraction accuracies of the two images NB33E020014 and NB34E017017 are the highest, and their qualities are 70.42% and 71.60% respectively. The surface cover differences between the valley areas and non-valley areas of the two images are significant. Among them, the vegetation distribution symmetry on the valley line of the NB34E017017 image is better. It should be noted that the DEM grid spacing used in the embodiments of the present invention is 10 meters, and this buffer constraint condition is relatively strict; 2) When the buffer radius is 2 pixels, the recall rate and precision rate of the extraction results of all images are higher than 87%, and the quality is higher than 77.8%, indicating that the accuracy (geometric accuracy, accuracy) of automatically extracting valley lines is relatively high.

[0094] Figure 6 This is the comparison between the valley lines extracted by the method in this paper and the manually collected valley lines. From Figure 6 It can be seen that the geometric positions of the automatically extracted results and the manually collected results are basically the same. The characteristic lines reflecting the important valley forms are all completely extracted and have good connectivity. There are a small number of mis-extraction and missed-extraction phenomena in the automatic extraction method, mainly manifested as the missed extraction of small branches near the main valley line, or the mis-extraction of the valley line caused by the mis-extraction of the vegetation surface. Therefore, a small amount of manual intervention is required to correct the errors.

[0095] In this embodiment, the manual collection and automatic extraction times of the characteristic lines of all trial production data are counted, and the results are as Figure 7 shown. It can be seen from the figure that the manual collection time of a single image is 2 to 4 times that of the automatic extraction method. Comparing the cumulative processing time of images, the difference in the total time required by the manual collection method and the automatic extraction method increases continuously with the increase in the number of processed images. Therefore, compared with the traditional production method, using the automatic extraction method to batch extract valley lines greatly improves the production efficiency and has obvious advantages.

[0096] Furthermore, the filtering effect of hilly vegetation using the automatically extracted characteristic lines is as Figure 8 shown. In the DSM results, the vegetation forms in the valley areas are obvious, and the landform forms of the valleys themselves are not reflected. Using the DEM results edited by the characteristic lines, the canopy elevations of shrubs and trees are reduced to the ground elevation, and the vegetation is basically completely filtered out, improving the noise phenomenon of the preliminary filtering results. At the same time, the DEM data can reasonably reflect the valley characteristics of different topographies and landforms in the non-survey area of Central Africa. In addition, it is found by checking in the stereoscopic environment that the contour lines at the valleys fit well with the stereoscopic model.

[0097] The quality of the DEM product is further evaluated using manually collected checkpoints. The checkpoints are collected by experienced operators in a stereoscopic environment, and their elevation values are measured using forest gaps or adjacent bare lands. 15 - 20 checkpoints are collected for each map, evenly distributed in the valleys within the map sheet. The quality assessment results of eight 1:50,000 standard - sized map data are as Figure 9 shown. It can be seen from the figure that the root - mean - square error of the elevation of the checkpoints in the first type of geomorphic form (i.e., the valleys are covered with dense arbors, and the slopes are covered with sparse grasslands or bare lands) is the smallest because the vegetation cover types in the valleys and on the slopes are significantly different, and the automatically extracted valley line positions are relatively accurate. The root - mean - square errors of the elevations of the second and fourth types are relatively small, ranging from 2.4 to 3.1 meters. A small number of checkpoints with large elevation differences are often distributed in the valley areas with sparse vegetation, caused by excessive height reduction, but the error values are significantly lower than the maximum error limit (12 meters) and will not cause the root - mean - square error of elevation (6 meters) to exceed the limit. For the third type of geomorphic form, the points with large elevation differences are basically distributed in areas where the vegetation cover types in the valleys and on the slopes are relatively similar and the patch fragmentation is large. As shown in the quality assessment results of NB33E012012, its root - mean - square error of elevation is 3.5 meters, and the maximum error is 5.9 meters. The automatically extracted valley results of this type of geomorphic form need to be post - processed and corrected, or the filtering results can be further optimized and edited in a stereoscopic environment.

[0098] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0099] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for extracting hilly terrain feature lines, characterized in that, The specific steps include: Construct a feature set and a sample library, and extract the vegetation coverage range at the valley; Use the distance transformation and curvature analysis methods to obtain the preliminary extraction result of the valley feature line; Use the topological analysis method to optimize the continuity of the feature line, and at the same time use the local spatial relationship and empirical knowledge to remove heterogeneous line segments to obtain the optimized valley feature line; The feature set includes texture features, elevation features, and spectral features, among which the elevation features and spectral features are respectively obtained through preprocessing according to the DSM and DOM result data; the sample library stores all vegetation samples in the vegetation area to be extracted; The specific steps for extracting the vegetation coverage range at the valley are as follows: Obtain the feature set required for remote sensing image classification, including spectral features, texture features, and elevation features, and perform band stacking to obtain a stacked image; Perform unsupervised segmentation on the stacked image to generate segmentation objects; Obtain the sample data required for remote sensing image classification; Use the feature set and samples as the input of supervised classification for object-oriented image classification, and then obtain the vegetation surface distribution map of the target area, and then extract the vegetation coverage range at the valley; The steps for extracting the valley feature line are as follows: Take the object-oriented image classification result as the input, use the Euclidean distance to calculate the minimum distance value from all pixels in the valley vegetation area to the range boundary to obtain distance field data; Use the curvature analysis method to take the derivative of the distance field data, calculate the gradient data, and then perform reclassification and masking operations on the gradient data to extract the gradient maximum value area to obtain the preliminary extraction result of the valley feature line; The specific calculation steps of the distance field data are as follows: Use the Euclidean distance to calculate the minimum distance value from all pixels within the vegetation coverage range at the valley in the binary classification map to obtain distance field data; De(x,y) = [(x2 - x1) 2 + (y2 - y1) 2 1 / 2 ;​ where x i is the longitude coordinate value, and y i is the latitude coordinate value; The specific steps of the curvature analysis method are as follows: Calculate the gradient magnitude S of the distance field data, and connect the points with larger gradient values to obtain a skeleton line; S(x,y,De) = [(d de / d x ) 2 +(d de / d y ) 2 1 / 2 ;​ where d de is the derivative of the distance field data, dx is the derivative in the longitude direction, and dy is the derivative in the latitude direction; For the continuity optimization of the valley feature line, introduce the raster expansion method to fill in missing values and alleviate the problem of multiple small closed-loop skeleton lines; furthermore, introduce the topological refinement method to iteratively strip the outermost pixels to extract a valley line with a single pixel width; perform buffer analysis on the valley line with a single pixel width; then introduce the strategy of hanging nodes to remove non-feature lines, and then obtain the optimized valley feature line.

2. A refined production method of hilly DEM using the hilly terrain feature line extraction method described in claim 1, characterized in that, The specific steps include: Filter and edit the DSM sheet result data, reduce the surface elevation to the ground elevation to obtain the preliminary DEM filtering result; Perform optimization processing on the preliminary filtering result. For the valley feature line, measure the vegetation height using the forest gap and assign the elevation attribute to the feature line, and perform secondary filtering and editing on the DEM data through the three-dimensional feature line to obtain the DEM result data with reasonable expression of geomorphic features.

Images