CN101980296A - Detection method of the main channel of the Yellow River based on spectral unmixing - Google Patents

Abstract

The invention discloses a spectral de-aliasing-based Yellow River mainstream line detection method, which is used for solving the technical problem of low accuracy in the conventional Yellow River mainstream line detection method. The technical scheme is that: the method comprises the following steps of: roughly segmenting the river according to spectral characteristics of a water area in a remote sensing image; extracting bank lines by utilizing edge detection and the like to determine a river area; segmenting the river according to the curve degree and direction of the river reach and the river distribution between south and north water edges; for the segmented riverway, extracting area de-aliasing-based mainstream points by utilizing the spectral and physical characteristics; and connecting the extracted mainstream points by adopting a multiscale analysis method to form a final mainstream line. The accuracy of the Yellow River mainstream line detection method is improved.

Description

Detection method of the main channel of the Yellow River based on spectral unmixing

technical field

The invention relates to a method for detecting the main channel of the Yellow River, in particular to a method for detecting the main channel of the Yellow River based on spectrum unmixing.

Background technique

The extraction of the main channel of the Yellow River is traditionally done manually. This method is not only time-consuming and laborious, but also easily affected by natural conditions such as weather. More importantly, it is difficult to grasp the changes of the main channel in time during the flood season. In recent years, remote sensing images have been widely used in water body identification, river extraction, water quality detection, flood detection, water area change detection and other fields. The methods used in these studies are basically based on the characteristics of water bodies in images and using traditional Several classification methods are used for detection. The continuous development of remote sensing technology and remote sensing image processing makes it possible to use remote sensing images to detect the main channel of the Yellow River. The main channel of the river channel is a prominent problem in the silted sandy bed channel with variable river regime. Since there are few international studies on the channel regime of sandy rivers, there is currently no information on the interpretation of the channel main channel and river regime from remote sensing images. Figure application research results report. Domestically, the current remote sensing image interpretation of the main river channel has only been initially applied in the Yellow River. However, image data processing and interpretation rely on manual visual inspection, and the whole process is labor-intensive and time-consuming. At the same time, visual interpretation based on remote sensing images is still difficult. The difficulty of accurately extracting the main channel of the river.

Contents of the invention

In order to overcome the deficiency of poor accuracy of the existing methods for detecting the main flow of the Yellow River, the present invention provides a detection method for the main flow of the Yellow River based on spectral unmixing, which performs rough segmentation of the river according to the spectral characteristics of the waters in the remote sensing image; Then use edge detection to extract the bank line and determine the river area; segment the river according to the curvature and direction of the river section and the river distribution between the north and south water edges; Spectral and physical characteristics extract the main slip points based on regional unmixing; the multi-scale analysis method is used to connect the extracted main slip points to form the final main slip line, which can improve the accuracy of the Yellow River main slip line detection method.

The technical solution adopted by the present invention to solve its technical problems: a method for detecting the main line of the Yellow River based on spectral unmixing, which is characterized in that it includes the following steps:

(a) Use supervised classification and matching methods for river segmentation, and perform morphological processing on the classified images according to the shape characteristics of the Yellow River reach, merge small areas, and eliminate smaller tidal flats and bridges;

(b) Use the Canny operator to detect the edges, track and connect the detected edges according to the neighborhood method, connect the preliminary waterlines on the north and south banks of the Yellow River, remove the interfering line segments, and obtain a group of useful line segments; then according to the statistical characteristics, The number of points judged in each line segment belonging to a certain bank is greater than a certain threshold, and the north and south bank images are obtained; the line segments on the north and south banks are stored in a certain order in a matrix, and the complete waterlines of the north and south banks are obtained;

(c) Transform the spatial dam bank into the curvature domain through the curvature calculation of the window, and the position of the maximum point of the obtained curvature sequence represents the position of the top of the curve; the minimum value between two consecutive extreme points The position of the point represents the position of the transition point between two continuous river bends, which divides the Yellow River into a typical reach and an atypical reach;

(d) Use the SMACC algorithm to extract pure endmembers, use the AMGS algorithm to find the pole, and save the pixel position with the highest brightness in q(1)

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 0</span>}}$

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}$

Second, remove the mapping of the currently selected vector from the existing dataset

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}$

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}\frac{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; )</span>}$

Repeat the above two steps until the residual meets a certain threshold; determine the position of the main slip point;

定义几何区域为R(j，k，l)为平行四边形区域，其水平宽度为w＝2^-j，垂直厚度为t＝2^j-J+r；其中j＝1...J；定义k为区域R水平的位置；定义变量L表示区域的倾斜程度；H为区域R的垂直位置；(e) ① Divide the original data space into geometric regions at multiple scales; the largest scale

Define the geometric area as R(j, k, l) as a parallelogram area, its horizontal width is w=2 ^-j , and its vertical thickness is t=2 ^j-J+r ; wherein j=1...J; define k is the horizontal position of the region R; define the variable L to represent the inclination of the region; H is the vertical position of the region R;

为划分的最大尺度；划分的每一个纵向区域的宽度为w＝2^-j，其中j＝1...J；每一个纵向区域左右边界上任意两点连接，用这条线作为中线分别在上下做两条平行线，得到一个平行四边形区域，设厚度为t＝2^j-J+r；此平行四边形区域即数据结构；用s定义中线的斜率；用h定义中线左端点的垂直位置；定义两个尺度因子，δ₁＝t/(Vw)，δ₂＝t/U，分别对应斜率和垂直位置的分辨率；将数据空间在每一个尺度下划分平行四边形区域的集合；定义区域倾斜的斜率的绝对值不超过S；将平行四边形区域表示为R(j，k，l，i)，其中分别为尺度，0≤k＜1/w^-1，-Sδ₁≤l≤Sδ₁，0≤i≤δ₂ ^-1-1；对于R(j，k，l，i)，左侧垂直边的横坐标为x＝kw，中线与左侧垂直边的交点为y＝iδ₂，斜率s＝lδ₁；Binary partition the two-dimensional space along the horizontal axis;

is the maximum dimension of the division; the width of each divided vertical area is w=2 ^-j , where j=1...J; any two points on the left and right borders of each vertical area are connected, and this line is used as the middle line respectively in Make two parallel lines up and down to get a parallelogram area, set the thickness as t=2 ^j-J+r ; this parallelogram area is the data structure; use s to define the slope of the center line; use h to define the vertical position of the left end of the center line; Define two scale factors, δ ₁ =t/(Vw), δ ₂ =t/U, respectively corresponding to the slope and the resolution of the vertical position; divide the data space into a set of parallelogram areas at each scale; define the area slope The absolute value of the slope of does not exceed S; the parallelogram area is expressed as R(j, k, l, i), where respectively scale, 0≤k<1/w ^-1 , -Sδ ₁ ≤ l≤Sδ ₁ , 0≤i≤δ ₂ ^-1 -1; for R(j,k,l,i), the abscissa of the left vertical side is x=kw, the intersection point of the middle line and the left vertical side is y=iδ ₂ , the slope s=lδ ₁ ;

Continuity condition between two parallelogram regions R ₁ (j ₁ , k ₁ , l ₁ , i ₁ ) and region R ₂ (j ₂ , k ₂ , l ₂ , i ₂ ):

●The two regions are on the same scale, that is, j ₁ =j ₂ ;

●The two areas are adjacent, that is, |k ₁ -k ₂ |=1;

●The distance between the common vertical edge and the intersection of the two areas is very close, that is, |l ₁ +i ₁ -l ₂ |<v;

●The slopes of the midlines of the two regions are not much different, that is, |l ₁ -l ₂ |<u;

Define the two areas that meet the above four conditions as areas with good continuity;

② Count the number of points in the geometric area at each scale, and select the significant area according to the threshold;

③Establish an undirected graph G _j = (V _j , E _j ) at each scale, and the geometric area when Count>N is used as the vertex v∈V _j of the graph; where Count is the number of points in the area R, N is a defined threshold within the region R;

If two geometric areas meet the condition of continuity, connect an edge e∈E _j between the two vertices; calculate the number of points in the geometric area one by one, make a choice according to the threshold N and establish an edge with the continuity relationship, get an undirected graph;

④Use the depth-first search algorithm in each established undirected graph to search for the longest path, and the longest path found is the main slip line.

The beneficial effect of the present invention is: due to the spectral characteristics of the waters in the remote sensing image, the rough segmentation of the river is carried out; the edge detection is used to extract the bank line to determine the river area; Segment the river according to the river distribution between them; use the spectral and physical characteristics of the main slip line to extract the main slip points based on regional unmixing; apply the multi-scale analysis method to analyze the extracted main slip points Connect to form the final main slip line, which improves the accuracy of the detection method of the Yellow River main slip line.

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

Description of drawings

The accompanying drawing is a flow chart of the method for detecting the main channel of the Yellow River based on spectral unmixing in the present invention.

Detailed ways

Refer to attached picture. 1. River division.

As input a TM multispectral image, the first is a rough segmentation of the river. Spectral classification and matching techniques, such as spectral vector matching, Mahalanobis distance segmentation and Gauss Markov, are used to classify spectral images, and post-classification processing is carried out according to the shape characteristics of the Yellow River reach, such as region merging. The supervised classification method was used in the classification. Due to the influence of weather and other factors on the image, the spectra of the upper and lower reaches of the Yellow River in the image are quite different. If a single sample is used for spectral angle classification, there will be a large error. Therefore, two samples are used. Classify the images individually and then synthesize the images. Due to the influence of bridges and tidal flats in the image, it is not convenient to extract the waterlines on both sides of the north and south banks. When the edge accuracy requirements are suitable for further research, the image is processed morphologically, that is, the image is dilated and eroded to eliminate the smaller ones. Beaches and bridges.

2. River bank extraction.

The Canny operator is used to detect the edges, and the detected edges are tracked and connected, and the water edges on the north and south banks of the river are classified to prepare for the next step of river segmentation. In order to facilitate the extraction of the north and south banks, the obtained edge lines need to be connected and tracked; according to the neighborhood method, a group of continuous line segments can be obtained by using an edge tracking algorithm, and some shorter interfering line segments can be removed according to the length. The shape characteristics of the shoals are removed to obtain a set of useful line segments. According to the distribution characteristics of the water body of the Yellow River, it is observed that it has certain horizontal and vertical characteristics. Based on this, an algorithm can be designed to judge the north and south banks of each line segment. Based on the consideration of algorithm efficiency, it is only necessary to judge the north and south shores of some representative points in each section to obtain the information of which shore the entire waterside line belongs to. According to statistical characteristics, if the number of judged points belonging to a certain bank in each segment is greater than a certain threshold, it can be judged whether the segment belongs to the south bank or the north bank, otherwise, it belongs to the other bank, thus obtaining the image of the north and south banks. According to the needs, the line segments of the north and south banks are stored in the matrix in a certain order, so as to obtain the complete waterside line on both sides of the bank, which is convenient for the extraction of the main slip line in the next step.

3. River segmentation.

The Yellow River is divided into typical and atypical sections in terms of the main channel of the Yellow River. Typical river sections are divided into three types: straight and slightly curved, curved and branched.

The shape of each type of river section is very different. The main way to describe the difference is to use the bending coefficient and curvature to describe the bending degree and direction of the river section, and the second is to determine whether there is river distribution between the north and south waterlines. branch. The spatial connection between river sections is also one of the basis of judging the main slip line of the Yellow River. Therefore, it is also very important for the calculation of the spatial distribution relationship between the river sections. The present invention defines the spatial distribution relationship as the connection of continuous bends River sections such as bends and bends, etc., are separated from the space by using the continuous change arrangement between the river sections. The specific segmentation method is as follows:

a) Transform the spatial dam bank into the curvature domain through the curvature calculation of the window, and the position of the maximum value point of the obtained curvature sequence represents the position of the bend top of the curve.

b) The position of the minimum point between two consecutive extreme points represents the position of the transition point between two continuous river bends.

4. Extract the main slip point.

Master slip extraction method based on region unmixing.

Shield the background outside the river channel, perform PCA transformation on a certain river section, and take the transformed local area for unmixing. Use the sequential maximum angle convex cone (SMACC) algorithm, set the number of end members to four, and perform pure end member extraction: first, use the poles to determine the convex cone, and use this to define the first end member Yuan. Afterwards, an oblique projection satisfying the above constraints is applied in the existing cone to generate the next end member. Adding cones creates new endmembers. This method is repeated until an endmember is generated that satisfies a certain tolerance, that is already in the convex cone, or until the specified number of endmember classes is satisfied.

When determining the pole, the successive orthogonalization algorithm AMGS (adaptation of the augmented modified Gram Schmidt) is used to find the pole. First, select the vectors from H that you want to keep. The initial vector can be selected randomly or according to certain criteria. Here, the pixel with the highest brightness is selected, and its position is stored in q(1).

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 0</span>}}$

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}$

Second, the mapping of the currently selected vector is removed from the existing dataset.

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}$

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}\frac{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; )</span>}$

Repeat the above two steps until the residual meets a certain threshold.

When determining the abundance image, each substance is represented by a subspace covering its variation, and the content of each substance in the end member is obtained by projecting onto this subspace. Using the characteristic that the pole vector cannot be represented by the non-negative linear combination of other points in the image, but the non-pole can, the subspace covered by each end member is found to determine the abundance. Among the abundance images corresponding to various endmembers, the abundance images corresponding to the endmember spectra with monotonically increasing characteristics are selected. After determining the abundance image, use the section maximum abundance positioning method to locate the main slip point, that is, perform extreme value analysis on the pixel values of each section of the river (the river section is approximated by a vertical line) to find the extreme point, That is, the point with the highest brightness in the abundance image, so as to determine the position of the main slip point.

5. The main point of connection.

In the above method of extracting main points, it is necessary to connect the obtained main points to obtain a line segment. The present invention adopts the curve connection method of multi-scale analysis:

Firstly, the main point data space is divided into various scales, and for each divided area, the number of points in the area is counted. Then build the undirected graph. Finally find the longest path in the graph. The specific steps are:

定义几何区域为R(j，k，l)为平行四边形区域，其水平宽度为w＝2^-j，垂直厚度为t＝2^j-J+r。其中j＝1...J。宽度和厚度取决于尺度的大小和数据点的数目。定义k为区域R水平的位置。另外定义变量L表示区域的倾斜程度。H为区域R的垂直位置。a) Multi-scale division: Multi-scale division is performed on the original data space and divided into geometric regions. First, define multi-scale geometric regions with different orientations, scales, and lengths. Maximum scale

The geometric region is defined as R(j, k, l) as a parallelogram region with a horizontal width of w=2 ^-j and a vertical thickness of t=2 ^j-J+r . where j=1 . . . J. Width and thickness depend on the size of the scale and the number of data points. Define k to be the location of the region R level. In addition, the variable L is defined to represent the slope of the region. H is the vertical position of the region R.

为划分的最大尺度。划分的每一个纵向区域的宽度为w＝2^-j，其中j＝1...J。每一个纵向区域左右边界上任意两点连接，用这条线作为中线分别在上下做两条平行线，得到一个平行四边形区域，设厚度为t＝2^j-J+r。此平行四边形区域即为我们所需要的数据结构，通过平移和旋转中线得到更多的平行四边形区域。使用s定义中线的斜率。并h定义中线左端点的垂直位置。另外定义两个尺度因子，δ₁＝t/(Vw)，δ₂＝t/U，分别对应斜率和垂直位置的分辨率。将数据空间在每一个尺度下划分平行四边形区域的集合。定义区域倾斜的斜率的绝对值不超过S。将平行四边形区域表示为R(j，k，l，i)，其中分别为尺度，0≤k＜1/w^-1，-Sδ₁≤l≤Sδ₁，0≤i≤δ₂ ^-1-1。因此对于R(j，k，l，i)，左侧垂直边的横坐标为x＝kw，中线与左侧垂直边的交点为y＝iδ₂，斜率s＝lδ₁。Then divide the two-dimensional space into binary along the horizontal axis.

is the maximum size of division. The width of each divided longitudinal region is w=2 ^-j , where j=1...J. Connect any two points on the left and right borders of each vertical area, and use this line as the center line to draw two parallel lines at the top and bottom respectively to obtain a parallelogram area with a thickness of t=2 ^j-J+r . This parallelogram area is the data structure we need, and more parallelogram areas can be obtained by translating and rotating the midline. Use s to define the slope of the median line. And h defines the vertical position of the left endpoint of the midline. In addition, two scale factors are defined, δ ₁ =t/(Vw), δ ₂ =t/U, corresponding to the resolution of the slope and the vertical position, respectively. Divide the data space into a collection of parallelogram regions at each scale. The absolute value of the slope defining the slope of the region does not exceed S. Denote the parallelogram region as R(j,k,l,i), where are scales, 0≤k<1/w ^-1 , -Sδ ₁ ≤l≤Sδ ₁ , 0≤i≤δ ₂ ^-1 - 1. Therefore, for R(j, k, l, i), the abscissa of the left vertical side is x=kw, the intersection point of the middle line and the left vertical side is y=iδ ₂ , and the slope s=lδ ₁ .

Define the continuity condition between two parallelogram regions R ₁ (j ₁ , k ₁ , l ₁ , i ₁ ) and region R ₂ (j ₂ , k ₂ , l ₂ , i ₂ ).

● Both regions are on the same scale, ie j ₁ =j ₂ .

● Two areas are adjacent, ie |k ₁ -k ₂ |=1.

●The distance between the common vertical edge and the intersection point of the two regions is very close, that is, |l ₁ +i ₁ -l ₂ |<v.

●The slopes of the midlines of the two regions are not much different, that is, |l ₁ -l ₂ |<u.

The two areas satisfying the above four conditions are defined as areas with good continuity.

b) Data statistics: count the number of points in the geometric area at each scale, and select significant areas according to the threshold.

c) Construct an undirected graph: Construct an undirected graph by using the significant region and the continuity of the region.

Create an undirected graph G _j =(V _j , E _j ) at each scale, use each of the above-mentioned geometric regions as the vertex v∈V _j of the graph, but do not establish a vertex for each geometric region. Instead, it is necessary to select the ones that meet the requirements, which are called candidate regions. The selection of the candidate area is determined by the number of data points in the area. Let the number of points in the region R be Count, and define a threshold N. When Count>N, this region is the candidate region.

If two geometric areas satisfy the condition of continuity, connect an edge e∈E _j between these two vertices. Calculate the number of points in the geometric area one by one, choose according to the threshold N, and then establish edges according to the continuity relationship, so as to establish an undirected graph.

Search for the longest path: use the depth-first search algorithm in each established undirected graph to search for the longest path, and the longest path that is searched out is the main slip line.

Claims (1)

1. a method for detecting the main slip line of the Yellow River based on spectral unmixing, characterized in that it may further comprise the steps: (a) Use supervised classification and matching methods for river segmentation, and perform morphological processing on the classified images according to the shape characteristics of the Yellow River reach, merge small areas, and eliminate smaller tidal flats and bridges; (b) Use the Canny operator to detect the edges, track and connect the detected edges according to the neighborhood method, connect the preliminary waterlines on the north and south banks of the Yellow River, remove the interfering line segments, and obtain a group of useful line segments; then according to the statistical characteristics, The number of points judged in each line segment belonging to a certain bank is greater than a certain threshold, and the north and south bank images are obtained; the line segments on the north and south banks are stored in a certain order in a matrix, and the complete waterlines of the north and south banks are obtained; (c) Transform the spatial dam bank into the curvature domain through the curvature calculation of the window, and the position of the maximum point of the obtained curvature sequence represents the position of the top of the curve; the minimum value between two consecutive extreme points The position of the point represents the position of the transition point between two continuous river bends, which divides the Yellow River into a typical reach and an atypical reach; (d) Use the SMACC algorithm to extract pure endmembers, use the AMGS algorithm to find the pole, and save the pixel position with the highest brightness in q(1) ${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 0</span>}}$ ${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}$ Second, remove the mapping of the currently selected vector from the existing dataset ${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}$ ${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}}\frac{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; )</span>}$ Repeat the above two steps until the residual meets a certain threshold; determine the position of the main slip point; (e) ① Divide the original data space into geometric regions at multiple scales; the largest scale

Define the geometric area as R(j, k, l) as a parallelogram area, its horizontal width is w=2 ^-j , and its vertical thickness is t=2 ^j-J+r ; where j=1...J; define k as the area The horizontal position of R; define the variable L to indicate the inclination of the area; H is the vertical position of the area R; Binary partition the two-dimensional space along the horizontal axis; is the maximum dimension of the division; the width of each divided vertical area is w=2 ^-j , where j=1...J; any two points on the left and right borders of each vertical area are connected, and this line is used as the middle line to make up and down respectively Two parallel lines get a parallelogram area, set the thickness as t=2 ^j-J+r ; this parallelogram area is the data structure; define the slope of the center line with s; define the vertical position of the left end point of the center line with h; define two scale factors, δ ₁ =t/(Vw), δ ₂ =t/U, respectively corresponding to the slope and the resolution of the vertical position; divide the data space into a set of parallelogram regions at each scale; define the slope of the region slope The absolute value of does not exceed S; denote the parallelogram area as R(j, k, l, i), where are the scales, 0≤k<1/w ^-1 , -Sδ ₁ ≤l≤Sδ ₁ , 0≤ i≤δ ₂ ^-1 -1; for R(j, k, l, i), the abscissa of the left vertical side is x=kw, the intersection of the midline and the left vertical side is y=iδ ₂ , and the slope s= _lδ1 ; Continuity condition between two parallelogram regions R ₁ (j ₁ , k ₁ , l ₁ , i ₁ ) and region R ₂ (j ₂ , k ₂ , l ₂ , i ₂ ): ●The two regions are on the same scale, that is, j ₁ =j ₂ ; ●The two areas are adjacent, that is, |k ₁ -k ₂ |=1; ●The distance between the common vertical edge and the intersection of the two areas is very close, that is, |l ₁ +i ₁ -l ₂ |<v; ●The slopes of the midlines of the two regions are not much different, that is, |l ₁ -l ₂ |<u; Define the two areas that meet the above four conditions as areas with good continuity; ② Count the number of points in the geometric area at each scale, and select the significant area according to the threshold; ③Establish an undirected graph G _j = (V _j , E _j ) at each scale, and the geometric area when Count>N is used as the vertex v∈V _j of the graph; where Count is the number of points in the area R, N is a defined threshold within the region R; If two geometric areas meet the condition of continuity, connect an edge e∈E _j between the two vertices; calculate the number of points in the geometric area one by one, make a choice according to the threshold N and establish an edge with the continuity relationship, get an undirected graph; ④Use the depth-first search algorithm in each established undirected graph to search for the longest path, and the longest path found is the main slip line.

Images