A method for missing completion of road extraction of a remote sensing image based on road width reasoning

Abstract

The present application relates to the field of remote sensing image road extraction, and aims to provide a remote sensing image road extraction missing completion method based on road width reasoning, so as to solve the problem that the road extraction result is incomplete due to the influence of the coverings on both sides of the road in the road extraction based on remote sensing images, and achieve the goal of automatic and accurate road extraction. In order to solve the above technical problem, the specific steps include the following: first, obtaining a remote sensing image to be extracted, and performing road extraction based on a U-net network; then, determining the position of the road intersection based on the road network data of the remote sensing image; second, performing corner point detection on the road extraction result based on a corner point detection algorithm, and matching the detected corner points with the road intersection; finally, performing road width reasoning based on the matching result of the road intersection and the corner points, and completing the road extraction result. The present application fundamentally eliminates the influence of the coverings on both sides of the road on the road extraction, and achieves accurate road extraction.

Description

Technical Field

[0001] This field relates to road extraction from remote sensing images, specifically to a method for missing road extraction and completion based on road width inference in remote sensing images. Background Technology

[0002] Roads are the carriers of human activities and the vehicles used, serving as links between various activity locations or regions, and are of paramount importance in geographic information. Remote sensing images, with their large detection range and rapid data acquisition speed, are suitable for large-scale environmental perception, leading to numerous studies on road extraction methods based on remote sensing images. However, a challenge in road extraction is how to eliminate the influence of roadside obstructions (most commonly trees), as these obstructions can affect the completeness of road extraction during remote sensing image processing.

[0003] To address the above issues, current methods for road extraction from remote sensing images can generally be categorized into three types: The first type is based on deep learning methods. This method uses segmented remote sensing images as input and road extraction results as output. In this case, the network's loss function is typically improved, or an attention mechanism is added to achieve better extraction results. The second type is based on multi-source data. In this case, the data used is no longer limited to remote sensing images but generally includes road network data, GPS data, etc. Additional information is introduced through auxiliary data to improve the accuracy of road extraction. The third type involves first detecting road intersections in the remote sensing image and then extracting roads based on these intersections. In this method, the authors obtain road intersection regions and road branches through tensor voting at road intersections, and then use a semi-automatic annotation method combined with geodesic and template matching to achieve road extraction. This method creatively proposes a new approach to road extraction based on road intersection information; however, the post-processing methods used are complex, difficult to implement, and require manual assistance. Summary of the Invention

[0004] The purpose of this invention is to provide a method for missing road extraction and completion based on road width reasoning in remote sensing images, so as to solve the problem that the road extraction results are incomplete due to the influence of occlusions on both sides of the road in road extraction based on remote sensing images, and to achieve the goal of automated and accurate road extraction.

[0005] To address the aforementioned technical problems, this invention provides a method for extracting and completing missing roads in remote sensing images based on road width inference, specifically comprising the following steps:

[0006] Step 1: Acquire the remote sensing image of the road to be extracted, and perform road extraction based on the U-net network;

[0007] Step 2: Based on the road network data extracted from the remote sensing image, determine the road intersections, denoted as (x, y), and denote the number of roads converging at the intersections as n;

[0008] Step 3: Detect corner points at road intersections based on the extracted road results using a corner detection algorithm;

[0009] Step 4: Determine the relative position of the corner point to the road based on the same-side method, and group the corner points. Corner points located within the angle between the same two roads are grouped into the same group, for a total of n groups;

[0010] Step 5: Match the road intersections with the extracted corner points, matching n corner points;

[0011] Step 6: Infer the width of each road based on the road intersections and matched corner points;

[0012] Step 7: Build a buffer for the road network data based on the road width to generate the final road extraction results.

[0013] The specific process of step 5 is as follows:

[0014] Step 5.1: Traverse all corner point groups and calculate the Euclidean distance between each corner point and the road intersection point in each corner point group;

[0015] Step 5.2: Select the corner point with the smallest distance in each corner point group and match it with the road intersection. A total of n corner points are matched. The line connecting the matched corner point and the intersection is denoted as reference line l1, ..., l n The corresponding corner points are denoted as [(x1, y1), ..., (x... n y n )).

[0016] The specific process of step 6 is as follows:

[0017] Step 6.1: For each reference line, calculate its projected length on the two adjacent roads. Let the two adjacent roads be S and M, the reference line be l1, and the corresponding corner point be (x1, y1):

[0018] Step 6.2: Randomly select a point on road S, excluding the intersection, and denote it as (x). S y S ), calculate the projection length of reference line l1 onto road S:

[0019] On road M, randomly select a point other than the intersection and denote it as (x). M y M ), calculate the projection length of reference line l1 onto road M:

[0020] Step 6.3: Calculate the half-width of road S: The width of road S is

[0021] Calculate the half-width of road M: Then the width of road M is

[0022] Step 6.4: Return to step 6.1 until all road widths have been calculated.

[0023] Compared with the prior art, the present invention has the following advantages:

[0024] 1. This invention utilizes the characteristic that road intersections are not affected by obstructions, and accurately infers the road width based on intersection information, fundamentally eliminating the influence of obstructions on road extraction and achieving accurate road extraction.

[0025] 2. This invention eliminates the need for manual intervention in road width reasoning, thus achieving automated road extraction. Attached Figure Description

[0026] Figure 1 This is a flowchart of the remote sensing image road extraction and missing completion method based on road width reasoning provided by the present invention.

[0027] Figure 2 This is a schematic diagram of the corner point matching method for road extraction and missing completion in remote sensing images based on road width reasoning provided by the present invention.

[0028] Figure 3 This is a schematic diagram of the road width inference method for road extraction and missing information completion in remote sensing images based on road width inference provided by the present invention. Detailed Implementation

[0029] The following will clearly and completely describe the concept, technical advantages, and technical effects of the present invention in conjunction with the accompanying drawings and specific embodiments, so as to fully understand the purpose, features, and effects of the present invention. It should be noted that the specific embodiments described herein are only for explaining the present invention and do not limit the present invention.

[0030] See Figure 1 This invention first acquires a remote sensing image of the road to be extracted and performs road extraction based on the U-net network. Then, it determines the locations of road intersections based on the road network data of the remote sensing image. Next, it performs corner detection on the extracted road results using a corner detection algorithm and matches the detected corners with the road intersections. Finally, it performs road width inference based on the matching results of road intersections and corners to complete the road extraction results. The specific steps are as follows:

[0031] Step 1: Acquire the remote sensing image of the road to be extracted, and perform road extraction based on the U-net network;

[0032] Step 2: Based on the road network data extracted from the remote sensing image, determine the road intersections, denoted as (x, y), and denote the number of roads converging at the intersections as n;

[0033] Step 3: Detect corner points at road intersections based on the extracted road results using a corner detection algorithm;

[0034] Step 4: Determine the relative position of the corner point to the road based on the same-side method, and group the corner points. Corner points located within the angle between the same two roads are grouped into the same group, for a total of n groups;

[0035] Step 5: Match the road intersections with the extracted corner points, matching n corner points;

[0036] In implementation, step 5 involves matching road intersections with extracted corner points. The specific process is as follows:

[0037] Step 5.1: Traverse all corner point groups and calculate the Euclidean distance between each corner point and the road intersection point in each corner point group;

[0038] Step 5.2: Select the corner point with the smallest distance in each corner point group and match it with the road intersection. A total of n corner points are matched. The line connecting the matched corner point and the intersection is denoted as reference line l1, ..., l n The corresponding corner points are denoted as [(x1, y1), ..., (x... n y n )).

[0039] Step 6: Infer the width of each road based on the road intersections and matched corner points;

[0040] like Figure 2 and Figure 3 As shown, in implementation, step 6 infers the width of each road based on the road intersections and matching corner points. The specific process is as follows:

[0041] Step 6.1: For each reference line, calculate its projected length on the two adjacent roads. Let the two adjacent roads be S and M, the reference line be l1, and the corresponding corner point be (x1, y1):

[0042] Step 6.2: Randomly select a point on road S, excluding the intersection, and denote it as (x). S y S ), calculate the projection length of reference line l1 onto road S:

[0043] On road M, randomly select a point other than the intersection and denote it as (x). M y M), calculate the projection length of reference line l1 onto road M:

[0044] Step 6.3: Calculate the half-width of road S: The width of road S is

[0045] Calculate the half-width of road M: Then the width of road M is

[0046] Step 6.4: Return to step 6.1 until all road widths have been calculated.

[0047] Step 7: Build a buffer for the road network data based on the road width to generate the final road extraction results.

Claims

1. A method for missing completion of road extraction from a remote sensing image based on road width reasoning, characterized in that, The method comprises the following steps: Step 1: Obtain a remote sensing image to be road extracted, and perform road extraction based on a U-net network; Step 2: Determine a road intersection point based on road network data of the extracted remote sensing image, denoted as (x, y), and the number of roads converging at the road intersection point is denoted as n; Step 3: Perform road intersection corner point detection on the road extraction result based on a corner point detection algorithm; Step 4: Determine the relative position of the corner point and the road based on a same-side method, group the corner points, and group the corner points located in the same two road angles into the same group, and there are n groups in total; Step 5: Match the road intersection point with the extracted corner point, and match n corner points; Step 6: Infer the width of each road based on the road intersection point and the matched corner point; Step 7: Perform buffer zone construction on the road network data based on the road width, and generate a final road extraction result; The specific process of step 5 is as follows: Step 5.1: Traverse all corner point groups, and calculate the Euclidean distance between the corner point in each corner point group and the road intersection point; Step 5.2: Select the smallest distance between each corner point group and the road intersection point for matching, a total of n corner points are matched, and the line connecting the matched corner points and the intersection point is recorded as the reference line l1,…,l n , and the corresponding corner points are recorded as [(x1,y1),…,(x n ,y n )].

2. The method of claim 1, wherein the method further comprises: The specific process of step 6 is as follows: Step 6.1: For each reference line, calculate the projection length of the reference line on the adjacent two roads, and let the adjacent two roads be S and M, the reference line be l1, and the corresponding corner point be (x1, y1): Step 6.2: Randomly select a point on the road S other than the intersection, denoted as (x S ,y S ), calculate the projection length of the reference line l1 on the road S: A point on the road M other than the intersection is randomly selected and recorded as (x M ,y M ), and the projection length of the reference line l1 on the road M is calculated: Step 6.3: Calculate the half-width of the road S: The width of the road S is then The half-width of the road M is calculated as: The width of the road M is then Step 6.4: Return to step 6.1 until all road widths are obtained.

Images