A road extraction method from high-resolution images based on extended road shape index
By using extended road shape index and multi-scale collaborative representation methods in high-resolution remote sensing images, the problem of single feature scales and inaccurate shape description in road extraction is solved, and a more accurate and robust road extraction effect is achieved.
Patent Information
- Application Number
- CN202110914630.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-08-10
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2041-08-10
AI Technical Summary
In the prior art, road extraction algorithms are inadequate in high-resolution remote sensing images due to the single road feature scale and inaccurate description of spatial shape features, which are difficult to effectively handle in complex scenarios.
Using a method based on the extended road shape index, the spectral and spatial autocorrelation characteristics of high-resolution images are extracted through the preprocessing module, and the spatial shape characteristics of the extended road area are calculated using superpixel segmentation method and region expansion algorithm, and combined with multi-scale collaborative representation. The road segmentation method is performed with the graph cutting method.
It effectively expands the descriptive scale of road spatial characteristics, improves the accuracy and robustness of road extraction, and overcomes the extraction difficulties caused by single scale and inaccurate shape characteristics in traditional methods.
Smart Images

Figure CN113627331B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of remote sensing target extraction, in particular to a high-resolution image road extraction method based on an extended road shape index. Background Art
[0002] Road network is an important part of transportation infrastructure and also an important basic data of geographical conditions. A complete, timely and accurate road network has a wide range of social needs in people's daily travel, e-commerce logistics distribution and national development strategic planning. It also plays an important role in social and economic activities. With the full arrival of the Internet era and the rapid development of mobile Internet and artificial intelligence technology, informatization and intelligence are the future development direction of cities. Taking smart city as an example, it is based on computer technology, multimedia technology, large-scale storage technology and Internet of Things technology, with broadband network as the link, using remote sensing, global positioning system, geographic information system, telemetry, simulation-virtualization and other technologies to describe the city in three dimensions with multiple resolutions, multiple scales, multiple time and space and multiple types. Digital city infrastructure is the foundation of smart city construction, and as an important part of it, the construction of high-precision three-dimensional road network is becoming more and more important. Due to the rapid development of aviation and aerospace remote sensing technology, remote sensing technology has gradually become an important tool for the rapid development of informatization in today's society. Remote sensing technology can quickly and accurately obtain a large amount of ground observation data. Compared with medium and low resolution satellite images, high-resolution remote sensing images have more massive and rich surface information, which provides more favorable conditions for the dynamic update of Geographic Information System (GIS) data, the development and application of GIS, and is also of great significance for map update, image matching, change detection, etc. Therefore, extracting roads from high-resolution images has become a current research hotspot.
[0003] At present, surveying and mapping data production departments mainly use manual visual interpretation methods to extract roads from remote sensing images and aerial images. The manual visual interpretation method is relatively easy to operate and has high extraction accuracy. The reason is that humans have a lot of prior information about roads, including road shape features, topological features, and contrast with other landforms. At the same time, manual visual interpretation is insensitive to image noise and can well overcome unfavorable factors such as buildings, shadows, and vegetation on both sides of the road that easily affect road extraction. Therefore, when human and time resources are sufficient, it is feasible to extract roads using manual visual interpretation methods; but for large areas and when road networks are required to be extracted quickly in emergency situations, the speed of manual extraction is far from meeting the requirements. At this time, it is necessary to use computers to realize automatic or semi-automatic extraction of road networks to reduce the workload of internal work and improve work efficiency. However, in practical applications, due to the complexity of roads, such as image noise, different road materials (cement, asphalt), buildings, vegetation, and shadows, the existing road extraction algorithms have not yet completely solved the road extraction problem. Therefore, extracting roads from remote sensing images is still a research difficulty in the field of remote sensing and needs further research. Summary of the invention
[0004] The purpose of the present invention is to provide a high-resolution image road extraction method based on an extended road shape index, aiming to solve the problem of the single scale used for road features in traditional road extraction methods in the prior art, and to solve the problem of inaccurate description of the spatial shape features of roads in traditional object-oriented road extraction methods.
[0005] To achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a high-resolution image road extraction method based on an extended road shape index, comprising the following modules; a preprocessing module: extracting the spectral characteristics and spatial autocorrelation characteristics of the high-resolution remote sensing image in a spot-by-spot manner, extracting image spots using a superpixel segmentation method, and calculating the spectral characteristics and spatial autocorrelation characteristics of each spot; an extended road shape index extraction module: after extracting the image spots, each spot is scanned and processed through an iterative program, and the extended area of each spot is obtained through a regional expansion algorithm, and the spatial shape characteristics of the extended road area are calculated, including a skeleton-based object linear index and an average road width feature; a road extraction module: estimating the road probability based on a multi-scale collaborative representation method, and finally using a graph cut method to perform road segmentation to obtain a road surface extraction result.
[0006] A further technical solution of the present invention is that, in the preprocessing module, the superpixel segmentation method includes the following steps: initializing seed points: uniformly distributing seed points in the image according to the set number of superpixels; reselecting seed points in the n*n neighborhood of the seed points, calculating the gradient values of all pixels in the neighborhood, and moving the seed points to the place where the gradient is the smallest in the neighborhood, where n=3; assigning class labels to each pixel in the neighborhood around each seed point; distance measurement: including color distance and spatial distance, for each searched pixel, calculating its distance from the seed point respectively; iterative optimization: the above steps are iterated continuously until the error converges, where the number of iterations is 10; enhancing connectivity: creating a new label table, the elements in the table are all -1, and discontinuous superpixels and superpixels with too small a size are reallocated to adjacent superpixels in a Z-shaped direction, and traversed superpixels are assigned to corresponding labels until all superpixels are traversed.
[0007] A further technical solution of the present invention is that the extended road shape index extraction module includes two parts: area expansion with the segmented pattern as the center and spatial feature extraction based on the extended area.
[0008] A further technical solution of the present invention is that the area expansion centered on the segmented spot includes the following steps: introducing the first law of geography to extract image spatial features, and using the Moran index to represent spatial correlation; superpixel shape features based on spectral features, spatial features, and texture features, according to the spatial extension characteristics of roads in high-resolution remote sensing images and the uniform regularity characteristics of the spots segmented using the SLIC superpixel segmentation method; spot area expansion algorithm steps: input a segmented image spot Oc, and output Or: a set of spots surrounding Or.
[0009] A further technical solution of the present invention is that the patch area expansion algorithm specifically includes the following steps: initialization step: add Oc to Or; collect the patches in contact with Oc in the topology in the container Ocon, where Ocon = O1, O2, O3, ... Ot; construct a feature vector Vc based on the spectral mean of the patch Oc, and use the same method to calculate the feature vector Vk of each in Ocon, where 1≤k≤t, compare the spectral distance between Vc and Vk, and select the nearest adjacent patch Os from Ocon; compare Os and Oc, if Os and Oc satisfy the constraints R1, R2 and R3, Os is accepted as a patch belonging to the same target source as Oc; add Os to Or, and at the same time Os replaces Oc for the next step of expansion exploration; from step 1 to step 5 is an iterative process, and when any of the three constraints is not satisfied, the iterative expansion terminates and returns to Or.
[0010] A further technical solution of the present invention is that the spatial feature extraction based on the extended area is: when the expansion iteration around the central pattern is terminated, a set of uniform and spatially continuous areas Or are output, and the skeleton-based object linear index and average road width are used to describe the spatial characteristics of the area Or composed of the pattern.
[0011] A further technical solution of the present invention is that the road extraction module process is as follows: road extraction based on multi-scale collaborative representation and graph cut algorithm, selecting a multi-scale collaborative representation method, using a linear combination of training samples to represent a test sample, obtaining the road probability of each object at a certain scale, all pixels in the same object are assigned a likelihood value the same as the object, all road probability maps of three different scales are fused into a whole, and finally the road possibility of each pixel is obtained; given an image I, an undirected graph is constructed, and a road probability map is obtained through a multi-scale collaborative representation method to segment the image.
[0012] The beneficial effects of the present invention are:
[0013] To address the problem that traditional road extraction methods use a single scale for road features, this paper extends the spatial feature extraction of roads to multiple scales, taking into account the multi-scale and spatial structure differences of roads in high-resolution remote sensing images. By constructing an extensible road shape index, the problem of inaccurate description of spatial shape features of roads in traditional object-oriented road extraction methods is solved. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 It is a flow chart of road extraction based on extended road shape index in the present invention. DETAILED DESCRIPTION
[0015] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings.
[0016] like Figure 1As shown in the figure, a method for extracting roads from high-resolution images based on an extended road shape index includes the following modules: a preprocessing module: extracting spectral features and spatial autocorrelation features of high-resolution remote sensing images in a spot-by-spot manner, extracting image spots using a superpixel segmentation method, and calculating spectral features and spatial autocorrelation features of each spot; an extended road shape index extraction module: after extracting image spots, each spot is scanned and processed through an iterative program, and the extended area of each spot is obtained through a region expansion algorithm, and the spatial shape features of the extended road area are calculated, including a skeleton-based object linear index and an average road width feature; a road extraction module: estimating road probability based on a multi-scale collaborative representation method, and finally using a graph cut method to perform road segmentation to obtain a road surface extraction result. It should be noted that: image spots can be extracted using a superpixel segmentation method based on a Simple Linear Iterative Clustering algorithm. The SLIC superpixel segmentation method is an object-oriented segmentation method. Compared with a single pixel, a spot has more available features, such as shape, size, and texture, and can smooth a lot of "salt and pepper" noise.
[0017] In this specific embodiment, in order to solve the problem that traditional road extraction methods use a single scale for road features, the spatial feature extraction of roads is extended to multiple scales in consideration of the multi-scale and spatial structural diversity of roads in high-resolution remote sensing images. In addition, an extensible road shape index is constructed to solve the problem of inaccurate description of spatial shape features of roads in traditional object-oriented road extraction methods.
[0018] In another specific embodiment of the present invention, in the preprocessing module, the superpixel segmentation method includes the following steps: initializing seed points, i.e., cluster centers: uniformly distributing seed points in the image according to the set number of superpixels; reselecting seed points in the n*n neighborhood of the seed points, calculating the gradient values of all pixels in the neighborhood, and moving the seed points to the place with the smallest gradient in the neighborhood, where n=3; the purpose of this is to avoid the seed points falling on the contour boundary with a large gradient, so as not to affect the subsequent clustering effect, and assigning a class label to each pixel point in the neighborhood around each seed point, i.e., which cluster center it belongs to; unlike the standard k-means search in the entire image, the search range of SLIC is limited to 2S*2S, which can accelerate the convergence of the algorithm , Distance measurement: including color distance and spatial distance. For each searched pixel, calculate its distance from the seed point respectively; Iterative optimization: The above steps are iterated continuously until the error converges, which can be understood as the cluster center of each pixel no longer changes. Experiments have found that 10 iterations can achieve ideal results for most images, so the number of iterations is generally 10, but not limited to 10 times, it can also be 8 times, 9 times, 11 times, 12 times, etc.; Enhance connectivity: Create a new label table, all elements in the table are -1, and reallocate discontinuous superpixels and superpixels with too small sizes to adjacent superpixels in a Z-shaped direction, from left to right and from top to bottom. Assign the traversed superpixels to the corresponding labels until all superpixels are traversed.
[0019] Specifically, the extended road shape index extraction module includes two parts: region expansion with the segmented patch as the center and spatial feature extraction based on the extended region.
[0020] Specifically, the region expansion with the segmented patch as the center includes the following steps: introducing the first law of geography to extract image spatial features, and using the Moran index to represent spatial correlation; superpixel shape features based on spectral features, spatial features, and texture features, according to the spatial extension characteristics of roads in high-resolution remote sensing images and the uniform regularity characteristics of patches segmented using the SLIC superpixel segmentation method; patch region expansion algorithm steps: input a segmented image patch Oc, and output Or: a set of patches surrounding Or.
[0021] Specifically, the patch area expansion algorithm specifically includes the following steps: initialization step: adding Oc to Or; collecting patches in contact with in the topology in a container Ocon, where Ocon = O1, O2, O3, ... Ot;
[0022] Construct the feature vector Vc based on the spectral mean of the spot Oc, and use the same method to calculate the feature vector Vk of each in Ocon, where 1≤k≤t, compare the spectral distance between Vc and Vk, and select the nearest adjacent spot Os from Ocon; compare Os and Oc, if Os and Oc satisfy the constraints R1, R2 and R3, then Os is accepted as a spot belonging to the same target source as Oc; add Os to Or, and replace Oc with Os for the next step of expansion exploration; from step 1 to step 5 is an iterative process, when any of the three constraints is not satisfied, the iterative expansion terminates and returns to Or.
[0023] Specifically, when the expansion iteration around the central patch is terminated, a set of uniform and spatially continuous regions Or is output, and the skeleton-based object linear index and average road width are used to describe the spatial characteristics of the regions Or composed of the patches.
[0024] Specifically, the process of the road extraction module is as follows: road extraction based on multi-scale collaborative representation and graph cut algorithm, using a multi-scale collaborative representation method, using a linear combination of training samples to represent a test sample, and obtaining the road probability of each object at a certain scale. All pixels in the same object are assigned a likelihood value that is the same as the object, and all road probability maps of three different scales are fused into a whole, and finally the road possibility of each pixel is obtained; given an image I, the GC algorithm constructs an undirected graph G = {υ, ε}, υ represents the set of pixels in the image, and ε represents the set of edges of the undirected graph between adjacent pixels. For the road extraction task, a label "1" is defined for the road category, and a label "0" is defined for the non-road category. GC attempts to minimize the objective function of the following regional terms and boundary terms. The road probability map is obtained through the above multi-scale collaborative representation method to segment the image.
[0025] In the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0026] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.
[0027] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.
Claims
1. A method for extracting roads from high-resolution images based on an extended road shape index, characterized by: Includes the following modules; Preprocessing module: extracts the spectral characteristics and spatial autocorrelation characteristics of high-resolution remote sensing images in a spot-by-spot manner, uses superpixel segmentation method to extract image spots, and calculates the spectral characteristics and spatial autocorrelation characteristics of each spot; Extended Road Shape Index Extraction Module: After extracting image patches, each patch is scanned and processed through an iterative procedure. The extended area of each patch is obtained through a region expansion algorithm. The spatial shape features of the extended road area are calculated, including the skeleton-based object linear index and the average road width feature. The extended road shape index extraction module includes two parts: region expansion with the segmentation patch as the center and spatial feature extraction based on the extended region; The region expansion with the segmented patch as the center includes the following steps: introducing the first law of geography to extract image spatial features and using the Moran index to represent spatial correlation; superpixel shape features based on spectral features, spatial features, and texture features, according to the spatial extension characteristics of roads in high-resolution remote sensing images and the uniform regularity characteristics of patches segmented using the SLIC superpixel segmentation method; the patch region expansion algorithm steps: input a segmented image patch Oc, output Or: a set of patches surrounding Or; The patch area expansion algorithm specifically includes the following steps: initialization step: add Oc to Or; collect the patches in contact with Oc in the topology into the container Ocon, where Ocon = O1, O2, O3, ... Ot; construct a feature vector Vc based on the spectral mean of the patch Oc, and use the same method to calculate the feature vector Vk of each in Ocon, where 1≤k≤t, compare the spectral distance between Vc and Vk, and select the nearest adjacent patch Os from Ocon; compare Os and Oc, if Os and Oc meet the constraints R1, R2 and R3, then Os is accepted as a patch belonging to the same target source as Oc; add Os to Or, and at the same time, Os replaces Oc for the next expansion exploration; from step 1 to step 5 is an iterative process, when any of the three constraints are not met, the iterative expansion terminates and returns to Or; road extraction module: estimate the road probability based on the multi-scale collaborative representation method, and finally use the graph cut method to segment the road to obtain the road surface extraction result; The road extraction module process is as follows: road extraction based on multi-scale collaborative representation and graph cut algorithm, using a multi-scale collaborative representation method, using a linear combination of training samples to represent a test sample, obtaining the road probability of each object at a certain scale. All pixels in the same object are assigned the same likelihood value as the object, and all road probability maps at three different scales are fused into a whole to finally obtain the road probability of each pixel. Given an image I, an undirected graph is constructed, and the road probability map is obtained through the multi-scale collaborative representation method to segment the image.
2. The method for extracting roads from high-resolution images based on an extended road shape index according to claim 1, wherein: In the preprocessing module, the superpixel segmentation method includes the following steps: Initialize seed points: distribute seed points evenly within the image according to the set number of superpixels; Reselect a seed point in the n*n neighborhood of the seed point, calculate the gradient values of all pixels in the neighborhood, and move the seed point to the place with the minimum gradient in the neighborhood, where n=3; Assign a class label to each pixel in the neighborhood around each seed point; Distance metric: including color distance and spatial distance. For each searched pixel, the distance between it and the seed point is calculated. Iterative optimization: The above steps are iterated continuously until the error converges, and the number of iterations is 10; Enhance connectivity: Create a new label table with all elements set to -1. Reassign discontinuous superpixels and superpixels that are too small to adjacent superpixels in a Z-shaped direction. Assign the traversed superpixels to the corresponding labels until all superpixels are traversed.
3. The method for extracting roads from high-resolution images based on an extended road shape index according to claim 1, wherein: The spatial feature extraction based on the extended area is as follows: when the expansion iteration around the central patch is terminated, a set of uniform and spatially continuous regions Or are output, and the skeleton-based object linear index and average road width are used to describe the spatial features of the region Or composed of the patches.
