CN109961065A - A method for target detection of surface ships - Google Patents

Abstract

The invention relates to a surface ship target detection method. The left, right, upper and lower boundaries of the region of interest are obtained by calculating the sum of the number of vertical edge points on the Sobel vertical edge map, so as to determine the region of interest; The method of combining segmentation and contour growth is used to segment the region of interest; the features of the suspected target are calculated on the grayscale image and the binary image corresponding to the region of interest, and the effective ship target is determined by feature discrimination. The invention has wide adaptability to scenarios and accurate detection rate, and has very important significance and value.

Description

A method for target detection of surface ships

technical field

The invention relates to the field of image target detection, in particular to a surface ship target detection method.

Background technique

Image object detection is an important branch of pattern recognition, which is widely used in the fields of transportation and national defense. As an important means of navigation and transportation, ships at sea have high economic and strategic value. Especially in the situation of local conflicts caused by the increasing tension of marine resources, controlling the target of ships is an important means to effectively and quickly control the conflict situation. At present, the detection of ships on the sea surface is mainly carried out in the sea-sky junction area in the image, and only the segmentation method is used to extract the ship target in the sea-sky junction area, thus limiting the scope of the target detection area, and the extracted ship target is incomplete, resulting in detection. The ship target is not found or the detected ship target is incomplete. Therefore, the method of target detection based on the combination of row-column mean segmentation and contour growth has important application value, and the method mainly involves image processing and pattern recognition.

SUMMARY OF THE INVENTION

In view of the deficiencies of the prior art, the present invention provides an accurate and effective method for detecting objects on the sea surface based on the combination of row-column mean segmentation and contour growth.

The technical scheme that the present invention adopts for realizing the above-mentioned purpose is:

A method for target detection of surface ships based on the combination of row-column mean segmentation and contour growth, comprising the following steps:

Step 1: Extract the region of interest by counting the number of vertical edge points on the Sobel vertical edge map;

Step 2: Segment the region of interest by using a segmentation method based on the combination of row and column mean segmentation and contour growth;

The method for extracting the region of interest is:

Step 1: Calculate the sum of the number of vertical edge points in each column on the Sobel vertical edge map to determine the left and right boundaries of the region of interest;

Step 2: Calculate the sum of the number of vertical edge points in each row on the Sobel vertical edge map to determine the upper and lower boundaries of the region of interest;

Step 3: The area surrounded by the left border, right border, upper border and lower border in the grayscale image is the region of interest.

The size of the infrared sea surface image is M×N. Calculate the sum of the number of vertical edge points in each column on the Sobel vertical edge map to determine the left and right boundaries of the region of interest, and count from the leftmost edge of the image to the right on the Sobel vertical edge map. The number of vertical edge points in each column and T _left (j), j=1, 2, ..., N, when T _left (j)>T ₀ and NumFirst _left =0, the corresponding value of T _left (j) The column position j is the left boundary _BL of the region; on the Sobel vertical edge map, count the number of vertical edge points and T _right (j) in each column from the rightmost end of the image to the left, j=N, N-1, .. ., 1, when T _right (j)>T ₀ and NumFirst _right =0, the column position j corresponding to T _right (j) is the right boundary _BR of the region; wherein, NumFirst _left is T _left (j) The number of times that the constraint condition T _left (j)>T ₀ is satisfied, and the initial value is 0; NumFirst _right is the number of times that T _right (j) satisfies the constraint condition T _right (j)>T ₀ , and the initial value is 0; T ₀ =5 . If _BL > _BR , then _BL= N/4 and _BR =N/2.

Calculate the sum of the number of vertical edge points in each row on the Sobel vertical edge map to determine the upper and lower boundaries of the region of interest. There are two cases: there are sea lines in the image and there are no sea lines in the image, and there are sea lines in the image. That is, the sea line detection is valid, and there is no sea line in the image, that is, the sea line detection is invalid. The first situation: when there is a sea line in the infrared sea surface image, it is defined by the left boundary _BL and the right boundary _BR of the region of interest. In the Sobel vertical edge graph, the extracted vertical edges are projected in the horizontal direction, that is, the sum of the number of edge points in each row is calculated; first, the number of edge points in each row is calculated from the bottom to the top from the position of the sea antenna. T _top (i), i =K, K-1, ..., 1, 1≤K≤M, when T _top (i)>0 and T _top (i-1)=0, the row position corresponding to i is the upper boundary B _T , calculate the number of edge points and T _down (i) in each row from the bottom to the top of the last row in the image, i=M, M-1,...,K, when T _down (i)=0 and T _down ( When i-1)>0, the row position corresponding to i is the lower boundary B _B ; the second situation: when there is no sea line in the image, at the Sobel vertical edge defined by the left and right boundaries of the region of interest In the figure, the number of edge points and T _top (i) in each row are calculated from top to bottom from the first row of the image, i=1, 2, ..., M, when T _top (i) > T _H and NumFirst _top =0, the row position corresponding to i is the upper boundary B _T , the number of edge points in each row and T _down (i) are calculated from the bottom to the top of the last row in the image, i=M, M-1,... , 1, when T _down (i)> _TH and NumFirst _down =0, the row position corresponding to i is the lower boundary B _B , where NumFirst _top is T _top (i) that satisfies the constraint T _top (i) > _TH times, the initial value is 0, NumFirst _down is the times that T _down (i) satisfies the constraint condition T _down (i)> _TH , and the initial value is 0; _TH =3. If B _T > _BB , then B _T =M/4, and B _B =M/2.

In the grayscale image, the area surrounded by the left border _BL , the right border _BR , the upper border _BT and the lower border _BB is the region of interest.

The described segmentation method based on the combination of row and column mean segmentation and contour growth is:

The size of the region of interest is M′×N′, and the segmentation method based on the row and column mean is used to segment the region of interest.

Row binarization, R(i, j) is the gray value of any point on the image after binarization of the region of interest, where i=1, 2, ..., M', j=1, 2, . .., N'; if (i, j) is a target point, then R(i, j)=1; if (i, j) is a non-target point, then R(i, j)=0.

Judging each non-target point R(i ₀ , j ₀ ) in the image after the binarization of the region of interest, where R(i ₀ , j ₀ )=0, if the binarized image of the region of interest is There is at least one target point in the left, right and downward directions of (i ₀ , j ₀ ): when j=1, 2, . . . , j ₀ -1, there is R(j ₀ , j)=1, when When j=j ₀ +1,...,N', there is R(i ₀ ,j)=1, when i=i ₀ +1,...,M', there is R(i,j ₀ ) -1; and the same target contour obtained by contour growth is searched in the left, right and bottom directions of the corresponding position in the binary contour image, then the current non-target point is set as the target point.

The present invention has the following beneficial effects and advantages:

1. The method of the present invention based on the combination of row-column mean segmentation and contour growth can obtain an accurate ship area, avoid the obtained ship area being too large or too small than the real ship, and provide the calculation of ship features and ship discrimination. Strong basis to improve the accuracy of ship detection;

2. The present invention does not have high requirements on the quality of the obtained infrared sea surface images, mainly the detection of infrared sea surface ship targets under complex backgrounds;

3. The present invention has wide adaptability to scenarios, accurate detection rate, and is of great significance and value.

Description of drawings

Fig. 1 is the method flow chart of the present invention;

Fig. 2 is the sea antenna detection process diagram of the present invention;

Fig. 3 is the sea antenna detection result diagram of the present invention;

Fig. 4 is the process diagram of region of interest extraction of the present invention;

Fig. 5 is the interest region segmentation process diagram of the present invention;

6 is a schematic diagram of a neighborhood background area of the present invention;

FIG. 7 is a graph of the detection result of the infrared sea surface ship target of the present invention.

Detailed ways

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

Figure 1 is a flow chart of the method of the present invention.

A method for detecting sea surface ships based on the combination of row and column mean segmentation and contour growth. First, the method based on Hough transform is used to detect sea lines; then, the left, right, top and bottom of the region of interest are calculated on Sobel vertical edge map The lower boundary is used to determine the region of interest; further, a method based on the combination of row and column mean segmentation and contour growth is used to segment the region of interest; next, the grayscale image and binary image corresponding to the region of interest are calculated. Features of suspected targets, and determine effective ship targets through feature discrimination.

1. Using the method based on Hough transform to detect sea lines, the specific steps are as follows;

Step 1: Convolve the infrared sea surface image with the vertical Sobel operator, and binarize the filtered image to obtain the Sobel horizontal edge image;

Vertical Sobel operator:

The second step: within the range of 10° and the accuracy is 0.5°, perform Hough transform on the Sobel horizontal edge image to obtain the transformation matrix, and obtain a straight line whose length should be greater than 160 pixels, and select the longest 3 lines as candidate lines; If the number of lines is greater than 160 pixels, all lines with a length greater than 160 pixels are selected as candidate lines; if there are no lines, there is no sea line in the image, and the sea line detection result is invalid.

Step 3: Calculate the effective length of the straight line, the method is:

S _dot is the number of valid points, LC _i is the number of consecutive points of the i-th line segment, and LI _j is the number of consecutive points of the j-th discontinuity.

Step 4: According to the absolute value of the gray mean difference of the 15 lines above and below the line, and the gray variance of the 15 lines below the line to determine whether all the candidate lines meet the constraints

and

If the constraints are met, the candidate line is a sea line, otherwise it is not a sea line, where,

_Nup =15, _Ndown =15, T _{i up} is the gray mean value of the ith row in the 15-line area on the straight line, and T _{i below} is the gray average value of the ith row in the 15-line area under the straight line, It is the average gray level of the 15 areas under the straight line.

Figure 2 shows the process diagram of the method based on the Hough transform of the present invention to detect sea lines, wherein (a) is the infrared sea surface ship image, (b) is the vertical Sobel operator filtering image, (c) is the Sobel horizontal Edge image, (d) is the sea line detection result.

Figure 3 shows the result of the sea line detection of the present invention; wherein (a) is the sea line detection effect in the case of sea surface stratification, (b) is the sea line detection effect under the backlight situation, and (c) is the sea line detection effect under the background of clouds .

2. Region of interest extraction

The first step: convolve the infrared sea surface image with the horizontal Sobel operator, and binarize the filtered image to obtain the Sobel vertical edge image;

Step 2: Calculate the left and right boundaries of the area, and count the number of vertical edge points in each column and T _left (j) from the leftmost edge of the image to the right on the Sobel vertical edge map, j=1,2,...,N, When T _left (j)>T ₀ and NumFirst _{left =} 0, the column position j corresponding to T _left (j) is the left boundary BL of the region; on the Sobel vertical edge map, count from the rightmost end of the image to the left The number of vertical edge points in each column and T _right (j), j=N, N-1, ..., 1, when T _right (j)>T ₀ and NumFirst _right =0, the value of T _right (j) is The corresponding column position j is the right boundary _BR of the region; wherein, NumFirst _left is the number of times that T _left (j) satisfies the constraint condition T _left (j)>T ₀ , and the initial value is 0; NumFirst _right is T _right (j ) the number of times that the constraint condition T _right (j)>T ₀ is satisfied, and the initial value is 0; T ₀ =5. If _BL >BR, then _BL= N/4 and _BR =N/2.

Step 3: Calculate the upper and lower boundaries of the area. There are two cases: there are sea lines in the image and there are no sea lines in the image. If there are sea lines in the image, the sea line detection is valid, and if there are no sea lines in the image, the sea line detection is effective. invalid. In the Sobel vertical edge map defined by the left boundary _BL and the right boundary _BR of the region of interest, the extracted vertical edges are projected in the horizontal direction, that is, the sum of the number of edge points in each row is calculated; the first case: when the infrared When there is a sea line in the sea surface image, the number of edge points in each row and T _top (i) are calculated from the bottom to the top from the position of the sea line, i = K, K-1, ..., 1, 1≤K≤M, when When T _top (i)>0 and T _top (i-1)=0, the row position corresponding to i is the upper boundary B _T , the number of edge points in each row and T _down are calculated from the bottom to the top of the last row in the image (i), i=M, M-1,...,K, when T _down (i)=0 and T _down (i-1)>0, the row position corresponding to i is the lower boundary B _B ; The second case: when there is no sea line in the image, calculate the number of edge points and T _top (i) in each row from top to bottom from the first row of the image, i=1, 2,...,M, When T _top (i)> _TH and NumFirst _top =0, the row position corresponding to i is the upper boundary B _T , and the number of edge points in each row and T _down (i) are calculated from the bottom to the top of the last row in the image. , i=M, M-1,...,1, when T _down (i)>T _H and NumFirst _down =0, the row position corresponding to i is the lower boundary B _B , where NumFirst _top is T _top (i) The number of times that the constraint condition T _top (i)> _TH is satisfied, the initial value is 0, and NumFirst _down is the number of times that T _down (i) the constraint condition T _down (i)> _TH is satisfied, and the initial value is 0; _TH =3. If B _T > _BB , then B _T =M/4, and B _B =M/2.

In the grayscale image, the area surrounded by the left border _BL , the right border _BR , the upper border _BT and the lower border _BB is the region of interest.

Figure 4 shows the process of extracting the region of interest of the present invention; wherein (a) is the infrared sea surface ship image, (b) is the horizontal Sobel operator filtering image, (c) is the Sobel vertical edge image, (d) ) to extract results for the region of interest.

3. Region of Interest Segmentation

Step 1: Binarize the region of interest by using the segmentation method based on the mean of rows and columns. The size of the region of interest is M′×N′, and R(i, j) is any point on the image after the binarization of the region of interest. , where i=1,2,...,M',j=1,2,...,N'; if (i,j) is the target point, then R(i,j)= 1; if (i, j) is a non-target point, then R(i, j)=0.

Step 2: Judging each non-target point R(i ₀ , j ₀ ) in the image after binarization of the region of interest, where R(i ₀ , j ₀ )=0, if the binarization in the region of interest There is at least one target point in the left, right and bottom directions of (i ₀ , j ₀ ) in the transformed image: when j=1, 2, . . . , j ₀ -1, there is R(i ₀ , j) =1, when j=j ₀ +1,...,N', there is R(i ₀ ,j)=1, when i=i ₀ +1,...,M', there is R(i , j ₀ )=1; and the same target contour obtained by contour growth is searched in the left, right and bottom directions of the corresponding position in the binary contour image, then the current non-target point is set as the target point.

Note: In order to make the segmentation results more accurate, a certain background area is reserved during implementation, and 20 pixels are expanded around the area of interest.

As shown in FIG. 5 is the process diagram of the segmentation process of the region of interest of the present invention; wherein (a) is the region of interest image, (b) is the result image of the row and column mean segmentation, (c) is the contour image, and (d) is based on the row and column mean value. Image results of the segmentation method combined with contour growing.

4. Suspected target calculation

Compute the suspected target geometric features in the region of interest segmented image.

Step 1: Use the method of connected region marking to mark the suspected target in the binary image of the region of interest, and record the edge pixel coordinates and quantity of each marked target at the same time;

Step 2: The centroid, area, width and height features of the suspected target can be obtained by calculation; then, all the suspected targets in the binary image are sorted in descending order according to the area of the suspected target.

Calculate the average gray level of the suspected target and the neighborhood background in the area of interest image, where the neighborhood background is the area surrounded by 3 circles around the minimum circumscribed rectangle of the target, as shown in FIG. 6 is a schematic diagram of the neighborhood background area of the present invention , the solid line frame is the minimum circumscribed rectangle of the target.

5. Suspected target identification

The suspected target needs to be identified as a valid ship target through feature discrimination, and the feature constraints are as follows:

a. Area: The total number of target pixels of the suspected ship.

The area of the valid target is >100 pixels.

b. Aspect ratio: The width pixel value of the minimum circumscribed rectangle of the suspected ship target is divided by the height pixel value.

A valid target has an aspect ratio between 0.5 and 10.

c. Contrast: the difference between the average brightness of the suspected ship target and the average brightness of the neighborhood background, where the neighborhood background is the average gray value of the area surrounded by the 3 circles around the minimum circumscribed rectangle of the target.

Contrast of valid targets > 5 (gray difference).

d. Position: the position coordinates of the smallest circumscribed rectangle of the suspected ship target.

Perform target discrimination on the suspected target to determine the real ship target, and obtain the target detection result.

FIG. 7 is a graph showing the result of infrared sea surface ship target detection according to the present invention.

Claims (5)

1. a kind of surface vessel object detection method, which comprises the following steps:

Step 1: extracting area-of-interest by calculating the quantity of vertical edge points on Sobel vertical edge figure；

Step 2: area-of-interest being split with the dividing method that profile growth combines using based on the segmentation of ranks mean value.

2. surface vessel object detection method according to claim 1, it is characterised in that: the extraction area-of-interest packet Include following procedure:

Step 1: on Sobel vertical edge figure calculate each column vertical edge points quantity and come determine area-of-interest a left side, Right margin；

Step 2: calculated on Sobel vertical edge figure every row vertical edge points number and come determine area-of-interest it is upper, Lower boundary；

Step 3: by left margin, right margin, coboundary and lower boundary area encompassed being region of interest in gray level image Domain.

3. surface vessel object detection method according to claim 2, it is characterised in that: described in Sobel vertical edge The quantity of each column vertical edge points is calculated on figure and includes: come the left and right boundary for determining area-of-interest

The size of infrared sea image is M × N；

Each column vertical edge points number and T are counted to the right from image left end on Sobel vertical edge figure_left(j), j=1, 2 ..., N, works as T_left(j) < T₀And NumFirst_leftWhen=0, T_left(j) the column position j corresponding to is the left margin in region B_L；

Each column vertical edge points number and T are counted to the left from the right end of image on Sobel vertical edge figure_right(j), j= N, N-1 ..., 1, work as T_right(j) > T0 and NumFirst_rightWhen=0, T_right(j) the column position j corresponding to is region Right margin B_R；

Wherein, NumFirst_leftFor T_left(j) meet constraint condition T_left(j) number of > T0, initial value 0； NumFirst_rightFor T_right(j) meet constraint condition T_right(j) > T₀Number, initial value 0, T₀=5；If B_L> B_R, then B_L=N/4, B_R=N/2.

4. surface vessel object detection method according to claim 2, it is characterised in that: described in Sobel vertical edge The number of every row vertical edge points is calculated on figure and includes: come the upper and lower boundary for determining area-of-interest

If sea horizon detection is effective, by area-of-interest left margin B there are when sea horizon in infrared sea image_L, the right Boundary B_RDefined by Sobel vertical edge figure, the vertical edge of extraction is made into horizontal direction projection, that is, calculates every row marginal point Number and；Firstly, calculating every row marginal point number and T from the bottom to top from sea horizon position_top(i), i=K, K-1 ..., 1,1≤ K≤M works as T_top(i) > 0 and T_top(i-1)=0 when, line position corresponding to i is set as coboundary B_T, the last line from image Every row marginal point number and T are calculated from the bottom to top_down(i), i=M, M-1 ..., K work as T_downAnd T (i)=0_down(i-1) 0 > When, line position corresponding to i is set as lower boundary B_B；

If sea horizon is not present in infrared sea image, sea horizon detection is invalid, by the left and right boundary institute of area-of-interest In the Sobel vertical edge figure of restriction, every row marginal point number and T are from top to bottom calculated from image the first row_top(i), i=1, 2 ..., M, works as T_top(i) > T_HAnd NumFirst_topWhen=0, line position corresponding to i is set as coboundary B_T, last from image A line calculates every row marginal point number and T from the bottom to top_down(i), i=M, M-1 ..., 1, work as T_down(i) > T_HAnd BumFirst_downWhen=0, line position corresponding to i is set as lower boundary B_B；

Wherein, NumFirst_topFor T_top(i) meet constraint condition T_top(i) > T_HNumber, initial value 0, NumFirst_down For T_down(i) meet constraint condition T_down(i) > T_HNumber, initial value 0；T_H=3；If B_T> B_B, then B_T=M/4, B_B=M/ 2。

5. surface vessel object detection method according to claim 1, it is characterised in that: described to be carried out to area-of-interest Segmentation includes following procedure:

Step 1: binaryzation being carried out to area-of-interest, obtains binary image；

Step 2: point non-targeted in binary image being judged, if the non-targeted point is left and right in binary image At least there is a target point with lower direction, and left and right, lower three directions of corresponding position are searched in two-value contour images To the same objective contour grown by profile, then current non-targeted point is set to target point；

Step 3: the non-targeted point of each of traversal binary image executes step 2.