A ROAD EXTRACTION METHOD BASED ON HIGH RESOLUTION REMOTE SENSING IMAGE

Aiming at the road extraction in high-resolution remote sensing images, the stroke width transformation algorithm is greatly affected by surrounding objects, and it is impossible to directly obtain high-precision road information. A new road extraction method combining stroke width transformation and mean drift is proposed. In order to reduce road holes and discontinuities, and preserve better edge information, the algorithm first performs denoising preprocessing by means of median filtering to the pre-processed image. Then, the mean shift algorithm is used for image segmentation. The adjacent parts of the image with similar texture and spectrum are treated as the same class, and then the fine areas less than the maximum stroke width are reduced. On the basis , the road information is extracted by the stroke width transformation algorithm, and the information also contains a small amount of interference information such as spots (non-road). In order to further improve road extraction accuracy and reduce speckle and non-road area interference, the basic operations and combinations in mathematical morphology are used to optimize it. The experimental results show that the proposed algorithm can accurately extract the roads on high-resolution remote sensing images, and the better the road features, the better the extraction effect. However, the applicability of the algorithm is greatly affected by the surrounding objects.


Introduction
With the improvement of the resolution of remote sensing images, the use of high-resolution images for road information extraction has important application value in car navigation, traffic management, urban planning, geographic information system database updating, and production of electronic maps. It is a hot topic for scholars at home and abroad. The appearance of high-resolution images, the more detailed presentation of the geometric spectral features of the road, which increases the possibility of accurately extracting road information from remote sensing images, but on the other hand, redundant details such as houses, barriers, shadows, vehicles All other factors will cause huge interference to road extraction, which in turn increases the difficulty of road extraction algorithm design [1] .
The existing methods of road extraction from high-resolution images can be basically divided into three categories. One is based on road region extraction method, which recognizes and extracts road information according to some characteristics of the region in a certain target, including mathematical morphology [2] , Hough transform [3] , etc. The second is based on the method of extracting the edge of the road. The main principle of the method is to detect the gray point of the image and mark it as an edge according to certain criteria.
Representative methods include edge detection operator method [4,5] , Snake model method [6] , etc. The third is based on road center line extraction method, which needs to manually set the initial direction and target point to track the road and extraction, its representative method is the template matching method [7,8] . Due to the influence of various factors in the imaging process of remote sensing image, The phenomenon that the same object presents different spectra and different objects present the same spectra exists in the image, which increases the use of a single algorithm to extract multiple types and is mostly used for text detection of natural scenes [11] .
There are few studies on road extraction. Hou Y Y [12] et al applied the stroke width transformation algorithm to the road extraction of UAV Image, and combined with the K-means clustering algorithm to achieve accurate road detection. Hailing Zhou [13]  SWT was originally introduced as a pre-processing step for text detection because the stroke width of text characters has a certain stability, while non-text characters have a very large change in stroke width, so the difference in stroke width between the two is Very high [14] . SWT is a partial image operation that calculates the most likely stroke width value for each pixel based on the color information of each pixel in the image [15] . The specific steps are as follows: 1) Establishing a width image having the same size as the original image, and setting the stroke width value of each pixel in the width image to infinity under initial conditions; 2) Performing edge detection on the image by using the canny operator to calculate the gradient direction of the edge point; 3) Starting from any edge point P, according to the gradient direction d p of the point, search for another edge point q corresponding to it along the ray r=p+n*dp(n>=0), if the gradient direction d q of the point q is found to be exactly , then, the spatial position distance between the point p and the point q, that is, the Euclidean distance 4) Each pixel in the above process may be given the stroke width value more than once, so each new value should be compared with the original value, taking a smaller value, that is, the stroke width of the final record of each pixel. The value is always the width of the thinnest stroke it corresponds to.

Mean Shift algorithm
The mean shift algorithm is a simple and effective iterative clustering method, which is a nonparametric method based on probability density gradient estimation. The core idea is to estimate the clustering behavior of feature space sample points, and search for the target position of the sample point in the feature space by iterative operation, that is, the position with the highest probability density, then move the sample point to the target position and repeat the iterative process. Until all sample points are included in a certain part [16] .
ChenY [17] extended the mean shift algorithm, which introduced the concept of kernel functions and bandwidth matrices. Let the sample set X be obtained in n-dimensional space d R by n-time independent sampling of the probability density estimation function   x fˆ, then: Where h is the bandwidth parameter, the kernel function The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3/W10, 2020 International Conference on Geomatics in the Big Data Era (ICGBD), 15-17 November 2019, Guilin, Guangxi, China

. Then the mean shift vector is
. Let the weights of the sampling points be equal,     , and the iterative formula for the mean shift is: The probability density at the weighted average However, it is found from the implementation of SWT algorithm that the extraction accuracy of SWT has a great dependence on the results of Canny edge detection. The Canny edge detector is sensitive to noise around the target object, and it is easy to detect many small non-road edges, and the interference is large. In addition, when the SWT is extracted from the road, the width of all the extracted information is smaller than the maximum stroke width value, in addition to the road information, there are non-road information smaller than the maximum stroke width value, such as green belts, trees, and buildings on both sides of the road, as well as some rivers, parking lots, hardened open spaces, etc., are highly likely to be mistaken for roads due to their similar characteristics to roads and widths less than the maximum stroke width value.
Therefore, using the SWT algorithm alone will inevitably extract non-road information and affect the road extraction accuracy.
In order to solve the above problems, effectively extract the The accuracy of road extraction is quantitatively evaluated by three indicators, namely completeness, accuracy, and quality [18] , expressed by CP, CR, and QL, respectively. The algorithm extraction effect of this paper is better than the direct SWT extraction. This is because the SWT algorithm is sensitive to background noise. It is easy to detect a lot of redundant information for the background complex image, and there is a big error extraction problem. Combine mean shift and SWT. It is a good way to make up for the above problems, and the extracted road edges are better, the burrs are less, and the road information is extracted accurately. It is proved that the

Conclusion
Based on the research of road extraction algorithm, this paper value of this value will further improve the automation of the algorithm, which will be the focus of the next step.