CN115082497A - SAR image road network extraction method considering angle texture features and POI - Google Patents

Abstract

The invention relates to an SAR image road network extraction method considering angle texture features and POI, which comprises the following steps: step 1: acquiring input data comprising SAR image data, binary segmentation data, POI point data and POI connection data; step 2: constructing a road network extraction model by considering the angle texture characteristics and the POI; and step 3: and (3) inputting the input data in the step (1) into the road network extraction model constructed in the step (2) to obtain a road network extraction result. Compared with the prior art, the method has the advantages of effective extraction of the road network in the SAR image, good extraction effect, good universality and the like.

Description

SAR image road network extraction method considering angle texture features and POI

Technical Field

The invention relates to the technical field of SAR image road network extraction, in particular to an SAR image road network extraction method considering angle texture characteristics and POI.

Background

Urban road network information is an important component of modern spatial information infrastructure. In the big data era, urban road network information is of great importance for high-precision road navigation and automatic driving. The method comprises the following steps of field measurement, geographic information acquisition, remote sensing and the like, and is an important means for acquiring urban road network information. Particularly, for supplement and update of road network information, ground and satellite remote sensing have unique advantages in the aspect of rapid road mapping. Satellite remote sensing is widely used for road extraction due to numerous data sources and abundant information, and optical images and synthetic aperture radar images are more applied. In the two, the SAR image has a great difficulty in extracting road information due to a special imaging mode. In particular, the road network extraction of the high-resolution SAR image is influenced by image noise and a complex structure of the road network, and thus a good result is difficult to obtain. In recent years, with the push of big data, a lot of geographic information is increasingly applied to mapping remote sensing. The POI data is one of the POI data, and not only has position information such as shopping malls, restaurants, scenic spots and the like, but also contains a large amount of road related information, and the information undoubtedly provides possibility for effective identification of roads.

Currently, there is little research on the use of POI data for road extraction, and in particular, there is little research on the use of SAR image road extraction. Therefore, for the road network extraction, it is necessary to study the combination of the SAR image road extraction and the POI information.

Disclosure of Invention

The invention aims to overcome the defects in the prior art and provide the SAR image road network extraction method which effectively extracts the road network in the SAR image, has good extraction effect and good universality and considers the angle texture characteristics and the POI.

The purpose of the invention can be realized by the following technical scheme:

an SAR image road network extraction method considering angle texture features and POI comprises the following steps:

step 1: acquiring input data comprising SAR image data, binary segmentation data, POI point data and POI connection data;

step 2: constructing a road network extraction model by considering the angle texture characteristics and the POI;

and step 3: and (3) inputting the input data in the step (1) into the road network extraction model constructed in the step (2) to obtain a road network extraction result.

Preferably, the method for acquiring the SAR image data comprises the following steps:

carrying out geometric correction on the SAR image through the track file and the corresponding DEM;

the method for acquiring the binary segmentation data comprises the following steps:

the corrected SAR image is subjected to binarization processing using a threshold segmentation method, and in a binary segmentation image, road, water, and shadow regions are divided into black regions having pixel values of 0, and other regions are divided into white regions having pixel values of 1.

Preferably, the method for acquiring POI point data comprises:

adopting POI classes of intersection names, using a rectangular frame to perform mobile calculation in a POI range, counting the number of points in the rectangular frame, if the number is more than 1, calculating the central positions of the points, only keeping the points closest to the central positions, and deleting other points; if the number of the points is 2, one of the points is reserved;

the calculation formula is as follows:

wherein (c) _x ,c _y ) Coordinates of the center position of the rectangular frame; (x) _Pi ,y _Pi ) Coordinates of each point in the rectangular frame; d _i The distance from each point to the central position; i ranges from 1 to n, n>2。

More preferably, the step 2 specifically includes:

step 2-1: establishing an initial template;

step 2-2: constructing a road identification submodel;

step 2-3: constructing a crossing identifier model;

step 2-4: and setting a moving strategy.

More preferably, the initial template in step 2-1 is a rectangular template.

More preferably, the road identifier model is specifically:

the rectangular template is defined as:

Template＝(α,M,V,S)

wherein alpha is the rotation angle of the rectangular template; m is a gray level mean value corresponding to alpha; v is the gray variance corresponding to alpha; s is road information in the binary image corresponding to the alpha;

when the rectangular template rotates, the following are corresponded to:

sequencing M in an ascending order, recording angles corresponding to a plurality of previous values, extracting V and S corresponding to the plurality of angles, and respectively sequencing in the ascending order; scoring a plurality of values of M, V and S, wherein the highest value is 1, and the rest values are reduced by 0.1 in sequence; after three sets of scores were obtained, the three were combined by the following formula and the new sets were sorted in descending order:

wherein, delta, gamma and omega are coefficients, and the value range is 0 to 1; a is the total score after combining the three types of information, and A' is the result of sequencing A in a descending order;

after sorting, selecting a plurality of directions with the highest total score as candidate directions, calculating the difference value of the gray average value between the current rectangular template and the next rectangular template, and recording the corresponding angles;

and selecting the direction with smaller difference value and smaller rotation angle as the advancing direction of the rectangular template by adopting a scoring method.

More preferably, the intersection identifier model is specifically:

identifying intersections based on POI points and POI connecting lines, judging the extending direction of the intersections after the POI points are found in the moving process of a rectangular template, when the POI connecting lines are consistent with the road direction, sequencing by applying the quantity of the POI connecting lines covering the road information, respectively obtaining the corresponding gray average value, gray variance and road information of a binary segmentation image, then performing fractional sequencing in a grading mode, and taking the direction with smaller gray average value difference and smaller rotation angle as the alternative direction for extending the intersections;

and when the POI connecting line direction is not consistent with the road direction, directly adopting a road identification submodel for judgment.

More preferably, the moving policy specifically includes:

and advancing the rectangular template by half the length of the template each time according to the road identification submodel.

More preferably, the steps 2 to 4 further include:

setting a stop movement strategy, and stopping movement when any one of the following conditions occurs:

reaching the edge of the image;

and the extracted road is reached.

Preferably, the road network extraction model is specifically:

the initial template is continuously advanced and judged as follows:

if the road reaches the extracted road, judging the candidate intersection;

when the candidate intersection exists and the corresponding road is not extracted, extracting the road by adopting a road identification sub-model;

otherwise, judging the next intersection until the extraction requirement is met;

if no candidate crossing exists or all the crossings are judged and do not meet the extraction requirement, the program is terminated;

if the extracted road is not reached, continuously judging whether the road intersection is reached; if the road intersection is judged, judging and storing according to the intersection identifier model; otherwise, continuing to judge whether the image edge is reached; if the candidate intersection is not found, the template direction is continuously judged;

the whole extraction process is the advancing process of the rectangular template, and the judgment is repeated every time until the strategy of stopping movement is met.

Compared with the prior art, the invention has the following beneficial effects:

firstly, effectively extracting a road network in an SAR image: the SAR image road network extraction method fully combines the SAR image angle texture characteristics and POI information, and effectively extracts roads while ensuring the integrity of the road network structure.

Secondly, the extraction effect is good: according to the SAR image road network extraction method, the gray mean value, the gray variance and the binary segmentation information of the SAR image are combined in a scoring mode, the problem that the measurement scales of different information are inconsistent is solved, and the extraction effect of a road network is effectively improved.

Thirdly, the universality is good: experiments prove that the SAR image road network extraction method can be used for extracting different types of road networks.

Drawings

FIG. 1 is a schematic flow chart of an SAR image road network extraction method in the present invention;

FIG. 2 is a schematic illustration of POI processing in an embodiment of the present invention;

fig. 2(a) is a schematic diagram illustrating a process of deleting duplicate POI data, and fig. 2(b) is a schematic diagram illustrating a process of determining a direction by connecting POI points;

FIG. 3 is a schematic diagram of an initial rectangular template in an embodiment of the present invention;

fig. 3(a) is a schematic diagram of the created initial rectangular template, and fig. 3(b) and fig. 3(c) are two cases after the initial rectangular template is adjusted, respectively;

FIG. 4 is a diagram of a road identification submodel in an embodiment of the invention;

FIG. 5 is a schematic diagram of an intersection identifier model according to an embodiment of the present invention;

wherein, fig. 5(a) is a schematic view of a rotation process of the rectangular template at the intersection, and fig. 5(b) is a schematic view of judgment of the template at the intersection;

FIG. 6 is a schematic diagram of an experimental area selected in an embodiment of the present invention;

FIG. 7 is a schematic representation of experimental area-I after POI treatment in an example of the present invention;

wherein, fig. 7(a) is a superimposed display diagram of POI links and POI points, fig. 7(b) is a POI link diagram of the experimental area-I, and fig. 7(c) is a POI point distribution diagram of the experimental area-I;

FIG. 8 is a schematic diagram of a road network extraction result according to an embodiment of the present invention;

wherein, fig. 8(a) corresponds to one intersection experiment, fig. 8(b) corresponds to two intersection experiments, and fig. 8(c) corresponds to three intersection experiments; FIG. 8(d) corresponds to four crossing experiments;

FIG. 9 is a diagram illustrating an abnormal region of the extraction result according to an embodiment of the present invention;

FIG. 10 is a schematic diagram of extraction results from other experimental regions according to an embodiment of the present invention;

wherein FIG. 10(a) is the extraction result of experiment region-II, and FIG. 10(b) is the extraction result of experiment region-III;

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, shall fall within the scope of protection of the present invention.

An SAR image road network extraction method considering angle texture features and POI, the flow of which is shown in FIG. 1, includes:

step 1: acquiring input data comprising SAR image data, binary segmentation data, POI point data and POI connection data;

step 2: constructing a road network extraction model by considering the angle texture characteristics and the POI;

and step 3: and (3) inputting the input data in the step (1) into the road network extraction model constructed in the step (2) to obtain a road network extraction result.

The following describes each step in detail:

data processing

(1) SAR image data:

carrying out geometric correction on the SAR image through the track file and the corresponding DEM;

(2) binary divided data

The corrected SAR image is binarized using a threshold segmentation method, and in a binary segmentation image, a road, water, and shadow area is divided into a black area having a pixel value of 0, and the other areas are divided into a white area having a pixel value of 1. In the road extraction process, the black area will be used as auxiliary information.

(3) POI point data

This embodiment uses POI class of intersection name. POI data is collected spontaneously by people and therefore has many duplicate messages. In the embodiment, the POI data are applied to identify intersections, so that duplicate data need to be removed. To achieve this, the present embodiment employs a filter-like approach. Within the POI range, a 100 × 100 meter rectangular box is used for the movement calculation. Counting the number of points in the rectangular box, and if the number is larger than 1, calculating the center positions of the points. Only the point closest to the center position is retained, and the other points are deleted. If the number of points is exactly 2, one of them is retained. The calculation formula is as follows:

wherein (c) _x ,c _y ) Coordinates of the center position of the rectangular frame; (x) _Pi ,y _Pi ) Coordinates of each point in the rectangular frame; d _i The distance from each point to the central position; i ranges from 1 to n, n>2。

Fig. 2(a) shows the processing of POIs, and the five-pointed star represents the central position of all POIs in the rectangular frame, and one point closest to the central position will be reserved. The processed POI points need to be buffered, because a smaller template rotation angle may miss the POI points when the POI point positions are inaccurate. The buffer area is made to increase the target range of the POI point, thereby facilitating the identification.

(4) POI connection data

The processed POI data only keeps one point at each intersection, so the points can be used for identifying the intersection. In actual roads, intersections are connected to each other, and if POI data can be effectively connected, the direction of the road can be determined. For a road, it can be generally considered as a continuous linear feature with a small curvature. In a dense road network, the direction of extension of a road intersection (except for an overpass or a special road) is not more than four in most cases. According to these features, the present embodiment connects POI data.

Assume that point a (in fig. 2 (b)) is the first point to be judged. And calculating the distance between the point a and other points according to the positions. And (4) sequencing the distances from small to large, and taking values in sequence. And sequentially calculating whether one point falls within an angle of-30 degrees to 30 degrees on the extension line of the connecting line between the point a and other points. If the direction exists, the straight line direction of the connection point a and the point is the initial direction, otherwise, the direction with the shortest distance is taken as the initial direction. It is then determined whether the angle between the other direction and the initial direction is greater than a threshold (e.g., 80). The direction with the larger angle is reserved until all directions are judged. And after the point a is removed, repeating the previous process, and sequentially judging the rest POI points. Each pair of points is connected, and the connection data is exported for buffer processing.

In fig. 2(b), α is an angle between ab and ac, and β is an angle corresponding to-30 ° to 30 °. The point indicated by the arrow falls within β, which is the initial direction. Points b, c, d and e are direction points that remain after the determination. POI links do not represent a true road network. Since the road is complicated, there may be a case of erroneous connection. However, the present embodiment is only to use the POI data to assist in determining the direction, so that a reasonable number of connections is kept as much as possible.

Second, model construction

The model construction process comprises the following steps:

step 2-1: establishing an initial template;

the initial template in this embodiment is rectangular, which has the advantage that the road width can be kept constant during the extraction process. The shape and position of the rectangular template are determined by the coordinates of three points using geometric relationships. The first point determines the direction of the road and is selected at the middle of the road. The second and third points are used to determine the bottom edge (width) of the template, selected on both sides of the road. In general, the length of the template should be greater than 2 times the width. The second point is located as accurately as possible at the edge of the road. To better fit the rectangular template to the road, after the three points are determined, the template will be adjusted centered on the second point. When creating a rectangular template, there are many cases due to different initial directions, and fig. 3(a) is one of them, and the corresponding formula is as follows:

X _n ＝X _C +S·sinα

Y _n ＝Y _C -S·cosα

X _m ＝X _n -BC·cosα

Y _m ＝Y _n -BC·sinα

wherein (X) _A ,Y _A ),(X _B ,Y _B ),(X _C ,Y _C ),(X _m ,Y _m )and(X _n ,Y _n ) Is the coordinate of each point in FIG. 3(a), and α is &. Fig. 3(b) and (c) illustrate two possible situations in the template adjustment process.

Step 2-2: constructing a road identification submodel;

in the SAR image, the gray level of the road is slightly lower than that of the surrounding ground objects, and in the binary segmentation image, the value of the road is generally 0. Based on this information, the computational content of the angular texture features is improved. The road extraction is mainly realized by combining the gray mean value, the gray variance and the binary segmentation information. The model assumes that the rectangular template is defined as:

Template＝(α,M,V,S)

wherein alpha is the rotation angle of the rectangular template; m is a gray level mean value corresponding to alpha; v is the gray variance corresponding to alpha; s is road information in the binary image corresponding to the alpha;

when the rectangular template rotates, the following are corresponded to:

sequencing M in an ascending order, recording angles corresponding to the first 10 values, extracting V and S corresponding to the 10 angles, and respectively sequencing in an ascending order; scoring 10 values of M, V and S, with the highest score of 1, and sequentially reducing the rest values by 0.1; taking M as an example, the scores of the top 10 minima are (1,0.9, …,0.1), respectively. Similarly, V and S also get a set of scores, which may be (0.9,1, …,0.2) and (1,0.9, …,0.3), respectively. It is normal that the score values of V and S are different from M because the ordering is different.

After three sets of scores were obtained, the three were combined by the following formula and the new sets were sorted in descending order:

wherein, delta, gamma and omega are coefficients, and the value range is 0 to 1; a is the total score after combining the three types of information, and A' is the result of sequencing A in a descending order; in the formula, δ, γ and ω are used to determine the weight of the three types of information, and may be modified according to the specific situation of the image, and the coefficients used in this embodiment are all 1. And after sorting, selecting the 5 directions with the highest total score as candidate directions. After sorting, selecting 5 directions with the highest total score as candidate directions, calculating the difference value of the gray average value between the current rectangular template and the next rectangular template, and recording the corresponding angles;

and selecting the direction with smaller difference value and smaller rotation angle as the advancing direction of the rectangular template by adopting a scoring method.

The main reason for this is that the reference quantities have different value ranges, and if they are simply added, the combination significance is lost. By adopting the mode in the embodiment, the sequencing of the reference quantities in the grouping is kept, and the importance of each value in the extraction process is considered. Fig. 4 takes a template as an example, and calculates the value of the template rotated by-30 degrees to-30 degrees. And selecting 10 angles with smaller gray mean values to obtain corresponding gray variance and binary segmentation image information, and then calculating the combined scores of the gray variance and the binary segmentation image information. And selecting 5 directions with the largest scores as candidates, calculating the difference of the gray level mean values of the templates, and selecting smaller difference values and rotation angles as advancing directions.

Step 2-3: constructing a crossing identifier model;

identifying intersections based on POI points and POI connecting lines, judging the extending direction of the intersections after the POI points are found in the moving process of a rectangular template, when the POI connecting lines are consistent with the road direction, sequencing by applying the quantity of the POI connecting lines covering the road information, respectively obtaining the corresponding gray average value, gray variance and road information of a binary segmentation image, then performing fractional sequencing in a grading mode, and taking the direction with smaller gray average value difference and smaller rotation angle as the alternative direction for extending the intersections;

since the intersections in this embodiment are mostly four-branch intersections, it is only necessary to judge the left and right sides of the traveling direction for the rectangular template, that is, the-50 ° to-130 ° and the 50 ° to 130 ° of the advancing direction of the template, as shown in fig. 5 (a). Although the POI link is buffered, it is determined before application due to the problems of connection error and inaccurate direction. And when the connecting line direction is consistent with the road direction, sequencing by using the quantity of the POI connecting line covering road information, respectively obtaining the corresponding gray mean value, gray variance and road information of the binary segmentation image, and then performing fractional sequencing. Similar to the road identifier model, the direction with smaller gray average value difference and rotation angle will be used as the alternative direction for extending the intersection. And when the POI connecting line direction is not consistent with the road direction, directly using a road identification method for judgment.

Step 2-4: and setting a moving strategy.

The mobile strategy specifically comprises the following steps:

and advancing the rectangular template by half the length of the template each time according to the road identification submodel. For roads, there are generally no large turns or abrupt angles. By utilizing the characteristic, the embodiment adds the criterion that the moving direction is kept unchanged when the gray average value is changed in a small range. That is, if the gray level mean value of the current rectangular template and the difference after twice advancing are smaller than the threshold, the direction is kept unchanged; otherwise, the road identification method is used to determine the heading direction.

The stop move strategy herein is divided into two cases: one is to reach the edge of the image; the other is to reach the already extracted link. Both of these situations require the determination of the remaining intersections. If the intersections still remain to be extracted, entering the next intersection for extraction; otherwise, the routine is terminated. In this process, it is necessary to determine whether the road corresponding to the intersection has been extracted. If the intersection is extracted, judging the next intersection until all intersections are identified; if not, the road extraction is directly carried out.

The road network extraction model in this embodiment is specifically:

the initial template is continuously advanced and judged as follows:

if the road reaches the extracted road, judging the candidate intersection;

when the candidate intersection exists and the corresponding road is not extracted, extracting the road by adopting a road identification sub-model;

otherwise, judging the next intersection until the extraction requirement is met;

if no candidate crossing exists or all the crossings are judged and do not meet the extraction requirement, the program is terminated;

if the extracted road is not reached, continuously judging whether the road intersection is reached; if the road intersection is judged, judging and storing according to the intersection identifier model; otherwise, continuing to judge whether the image edge is reached; if the candidate intersection is not found, the template direction is continuously judged;

the whole extraction process is the advancing process of the rectangular template, and the judgment is repeated every time until the strategy of stopping movement is met.

One specific experimental case is provided below:

selection of experimental area and data set

The experimental area is in city A, because the city expansion driven by road network construction is relatively fast. In the city center, the road is characterized in that: relatively narrow, covered by trees, and more tall buildings on both sides of the road. In urban edge areas, the roads are relatively wide and there are fewer trees. In summary, the road structure of city a is relatively curved and complex, and it is difficult to extract road network information. In the experiment, three areas are selected to test the road network extraction method, wherein an experimental area-I and an experimental area-III are located in an area a, and an experimental area-II is located in an area b, as shown in FIG. 6. Road network information was extracted as experimental data using the high-resolution three-dimensional SAR image data, as shown in table 1. In the experimental area-I, four ROADs form a "well-shaped" ROAD network, which is a typical ROAD network structure, and is used for testing the ROAD network extraction model provided by the invention, i.e., the ATP-ROAD model. The experimental area-II and the experimental area-III are used for verifying the universality of the ATP-ROAD model, wherein the experimental area-II is provided with a trunk ROAD with larger curvature and two branch ROADs on the same side; the road network structure of the experimental zone-III is irregular. The backscattering information of the SAR image road contains information of surrounding ground objects, so that the difficulty of road extraction is increased. Testing and validation in these complex areas will facilitate better application of the model in other areas.

TABLE 1 conditions of three experimental areas (high resolution three SAR images 1 m)

Famous city	Position of	Imaging mode	Breadth (Km)	Polarization mode	Range (pixels)
						Experimental area-I	Region a	Slider bunching	10×10	Single polarization	1643×1637
Experimental area-II	Region b	Slider bunching	10×10	Single polarization	911×1517
						Experimental zone-III	Region a	Slider bunching	10×10	Single polarization	875×1342

For POI, the POI is distributed on the road or on two sides of the road, and has obvious road correlation, which can provide help for road extraction. Generally, the categories of POIs are rich, and POI points of intersection names are selected as auxiliary information for road extraction. The POI data used in the experiment were from data published online in 2021 on a high-end map. There are 23 POI classes, including road facilities, traffic services, place names and addresses, etc. POI data of the intersection name is a subclass of the location name address information class. Only a few classes related to roads are shown in table 2 for explaining the subordinate condition of the intersection name POI data, and the POI classification code table of the high-grade map is referred to.

TABLE 2 affiliation of POI data related to roads

Second, POI data processing

Corresponding to experimental area-I, the POI range should be greater than the range of the area. The embodiment processes the POI data according to the methods of POI processing and POI connection, and the buffer area is 5 meters. The points in fig. 7(a) are POI points after buffer processing and deduplication calculation, and the lines are POI link lines after buffer processing. The links shown in the figures are erroneous, which explains why the links cannot be relied upon entirely at junctions to determine the direction of travel of the roadway. Fig. 7(b) shows the POI links corresponding to experimental region-i, and fig. 7(c) shows the POI points, both of which are used as input data in the experiment.

Three, different types of road network extraction results

In order to view the implementation process of the method in this embodiment, experiments at one intersection, two intersections, and three intersections were also performed, as shown in fig. 8. According to the method, the road network of the SAR image can be effectively extracted, and the extracted road at the intersection is relatively complete under each condition.

Fourth, analysis of experimental results

As can be seen from fig. 8, the extracted result in the present embodiment substantially matches the actual road. The results are represented as a single line due to the different widths of the four roads. The shape of the line is complete, and the intersection also meets the actual situation. In the experiment, 228 road points are automatically generated, points deviating from the road are used as error points to calculate, and the accuracy rate exceeds 70%, as shown in table 3.

TABLE 3 precision evaluation Table for Experimental zone-I

Extraction method	Manually extracting points	Automatic point number extraction	Number of error points	Rate of accuracy
					Artificial extraction	159	-	-	-
The invention	3	228	65	71.49％

In the extraction result, there are three abnormal regions, which correspond to the region-a, the region-B and the region-C in fig. 9, respectively, and the three abnormal regions are analyzed in this embodiment.

Area-a is a road intersection. Normally, the two left and right connecting lines of the region should be crossed at one point, but the experimental results do not cross at the same point. The retrospective extraction process finds that when the rectangular template is advanced upward along R1 (the initial position of the template is on R1, the position of the five-pointed star mark in fig. 9) to reach the intersection, only the left road extending direction is retained, and the right extending direction is discarded because the gray scale difference value is greater than the threshold value. The road on the right side is extracted through a template from the road intersection at the upper right corner. Therefore, this extracted road is not finally merged with the initially extracted left road according to the judgment. This is also demonstrated by the template map of region-A. In fact, this problem can connect the two by modifying the extraction policy. However, it is found that the R4 does not strictly meet at the intersection in the actual situation by looking at the optical image. Moreover, intersections in this case are not rare in real life, so it is considered that this case is not an extraction error, and therefore the extraction strategy is not modified.

region-B is a section of road R4. Compared with other extracted areas, the road extraction result of the area is relatively curved. After the optical image is checked, the shadow formed by the high-rise building in the section of the area is found. In the SAR image, this area is substantially mixed with the road. When the rectangular template enters the section of the area, the influence of the shadow area is generated, so that the phenomenon of the area-B is generated. Which is difficult to avoid. As can be seen from the template map, the rectangular template is adjusted as it passes through the area so that the following road is not affected by the area.

zone-C is near the top right corner road intersection of the image. The extraction result of the area has a convex point, so that the line type is not smooth enough. The reason for tracing the problem is because when the template advances from bottom to top along R3, the direction at which the intersection is judged is the direction in which it is projected in the figure, and is stored. When the direction enters road extraction as a candidate direction, the template is adjusted (initial direction is changed) quickly according to the method in the present embodiment, so the salient point problem of the region-C occurs. The problem is caused by two reasons, namely that the rectangular template is not flexible to rotate; secondly, the position of the POI point is inaccurate, and the judgment time of the extending direction of the intersection is influenced. Since this phenomenon is difficult to avoid, the extraction method will be improved subsequently to complete the situation.

Fifth, experimental results of other experimental zones

In addition to the experiment in experiment zone-I, this example also applied the ATP-ROAD method in two other zones, as shown in FIG. 10. Compared with the experimental area-I, the two areas have relatively complex road structures, but have smaller image ranges and clearer road surfaces, so that the extraction accuracy is higher, as shown in tables 4 and 5.

TABLE 4 accuracy assessment Table for Experimental area-II

Extraction method	Artificially extracting points	Automatic point number extraction	Number of error points	Rate of accuracy
					Artificial extraction	78	-	-	-
The invention	3	95	10	89.47％

TABLE 5 accuracy assessment Table for Experimental zone-III

Extraction method	Manually extracting points	Automatic point number extraction	Number of error points	Rate of accuracy
					Artificial extraction	110	-	-	-
The invention	3	141	25	82.27％

The application of POI data is a new attempt of a road extraction method, and many problems are encountered. Such as POI data redundancy issues. The present embodiment removes duplicate data using a filter-like approach. The POI connecting lines can identify road intersections and assist in judging the extending direction of roads to a certain extent. By using the POI data, the accuracy of the method for identifying the intersection is ensured.

In the road extraction process, it is difficult to maintain the integrity of the road network structure. The method fully combines the SAR image angle texture characteristics and the POI information, and effectively extracts the road while ensuring the integrity of the road network structure. The method combines the gray mean value, the gray variance and the binary segmentation information of the SAR image in a grading mode, and solves the problem that the measurement scales of different information are inconsistent. The experimental results of the three regions show that the method can be used for extracting different types of road networks.

While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and various equivalent modifications and substitutions can be easily made by those skilled in the art within the technical scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (10)

1. An SAR image road network extraction method considering angle texture features and POI is characterized by comprising the following steps:

step 1: acquiring input data comprising SAR image data, binary segmentation data, POI point data and POI connection data;

and 2, step: constructing a road network extraction model by considering the angle texture characteristics and the POI;

and step 3: and (3) inputting the input data in the step (1) into the road network extraction model constructed in the step (2) to obtain a road network extraction result.

2. The SAR image road network extraction method considering angle texture features and POI (point of interest) according to claim 1, characterized in that the SAR image data acquisition method comprises:

carrying out geometric correction on the SAR image through the track file and the corresponding DEM;

the method for acquiring the binary segmentation data comprises the following steps:

the corrected SAR image is subjected to binarization processing using a threshold segmentation method, and in a binary segmentation image, road, water, and shadow regions are divided into black regions having pixel values of 0, and other regions are divided into white regions having pixel values of 1.

3. The SAR image road network extraction method considering angle texture features and POI as claimed in claim 1, wherein the method for obtaining POI point data is as follows:

adopting POI classes of intersection names, using a rectangular frame to perform mobile calculation in a POI range, counting the number of points in the rectangular frame, if the number is more than 1, calculating the central positions of the points, only keeping the points closest to the central positions, and deleting other points; if the number of the points is 2, one of the points is reserved;

the calculation formula is as follows:

wherein (c) _x ,c _y ) Coordinates of the center position of the rectangular frame; (x) _Pi ,y _Pi ) Coordinates of each point in the rectangular frame; d _i The distance from each point to the central position; i ranges from 1 to n, n>2。

4. The SAR image road network extraction method taking the angle texture feature and the POI into consideration according to claim 1, wherein the step 2 specifically comprises:

step 2-1: establishing an initial template;

step 2-2: constructing a road identification submodel;

step 2-3: constructing a crossing identifier model;

step 2-4: and setting a moving strategy.

5. The SAR image road network extraction method considering angle texture features and POI as claimed in claim 4, wherein the initial template in step 2-1 is a rectangular template.

6. The SAR image road network extraction method considering angle texture features and POI as claimed in claim 4, wherein the road identification submodel is specifically:

the rectangular template is defined as:

Template＝(α,M,V,S)

wherein alpha is the rotation angle of the rectangular template; m is a gray level mean value corresponding to alpha; v is the gray variance corresponding to alpha; s is road information in the binary image corresponding to the alpha;

when the rectangular template rotates, the following are corresponded to:

sequencing M in an ascending order, recording angles corresponding to a plurality of previous values, extracting V and S corresponding to the plurality of angles, and respectively sequencing in the ascending order; scoring a plurality of values of M, V and S, wherein the highest value is 1, and the rest values are sequentially reduced by 0.1; after three sets of scores were obtained, the three were combined by the following formula and the new sets were sorted in descending order:

wherein, delta, gamma and omega are coefficients, and the value range is 0 to 1; a is the total score after combining the three types of information, and A' is the result of sequencing A in a descending order;

after sorting, selecting a plurality of directions with the highest total score as candidate directions, calculating the difference value of the gray average value between the current rectangular template and the next rectangular template, and recording the corresponding angles;

and selecting the direction with smaller difference value and smaller rotation angle as the advancing direction of the rectangular template by adopting a scoring method.

7. The SAR image road network extraction method considering angle texture features and POI as claimed in claim 4, wherein the intersection identifier model is specifically:

identifying intersections based on POI points and POI connecting lines, judging the extending direction of the intersections after the POI points are found in the moving process of a rectangular template, when the POI connecting lines are consistent with the road direction, sequencing by applying the quantity of the POI connecting lines covering the road information, respectively obtaining the corresponding gray average value, gray variance and road information of a binary segmentation image, then performing fractional sequencing in a grading mode, and taking the direction with smaller gray average value difference and smaller rotation angle as the alternative direction for extending the intersections;

and when the POI connecting line direction is not consistent with the road direction, directly adopting a road identification submodel for judgment.

8. The SAR image road network extraction method considering angle texture features and POI as claimed in claim 4, wherein the moving strategy is specifically as follows:

and advancing the rectangular template by half the length of the template each time according to the road identification submodel.

9. The SAR image road network extraction method considering angle texture features and POI as claimed in claim 4, wherein the steps 2-4 further comprise:

setting a stop movement strategy, and stopping movement when any one of the following conditions occurs:

reaching the edge of the image;

and the extracted road is reached.

10. The SAR image road network extraction method considering angle texture features and POI as claimed in claim 1, wherein the road network extraction model is specifically:

the initial template is continuously advanced and judged as follows:

if the road reaches the extracted road, judging the candidate intersection;

when the candidate intersection exists and the corresponding road is not extracted, extracting the road by adopting a road identification sub-model;

otherwise, judging the next intersection until the extraction requirement is met;

if no candidate crossing exists or all the crossings are judged and do not meet the extraction requirement, the program is terminated;

if the extracted road is not reached, continuously judging whether the road intersection is reached; if the road intersection is judged, judging and storing according to the intersection identifier model; otherwise, continuing to judge whether the image edge is reached; if the candidate intersection is not found, the template direction is continuously judged;

the whole extraction process is the advancing process of the rectangular template, and the judgment is repeated every time until the strategy of stopping movement is met.

Images