TWI736446B - Seabed edge detection method capable of filtering horizontal noise - Google Patents

Abstract

The present invention provides a seabed edge detection method capable of filtering horizontal noise, including: (a) input one Grayscale image step, (b) image pre-processing step, and (c) seabed edge detection and marking step; among them, (b) image pre-processing step includes: an image sharpening step, an image blurring step, and a step Otsu Binarization (Otsu Threshold) step; where (c) the seabed edge detection and marking step, according to the Otsu binarization side scan sonar image and through multiple sliding windows to find the seabed edge Line feature, the sliding window is used to detect the change of the pixel value in the Otsu binarized side scan sonar image. The advantageous effect that the present invention can achieve is that the seabed and water masses are separated through the (b) image pre-processing step, and the step of detecting and marking the edge of the seabed in (c) is to use a sliding window to detect Mark the location of the seabed edge.

Description

Seabed edge detection method capable of filtering horizontal noise

The invention belongs to a seabed edge detection method, in particular to a seabed edge detection method capable of filtering horizontal noise.

The side scan sonar system is an active sonar instrument that is widely used in underwater terrain and landform detection. Its principle is to transmit high-frequency sonar signals to the seabed through tow fish and receive the echo signals. The intensity is converted into a grayscale image of 0~255 (pixel, pixel). The side scan sonar system includes: fish towing device, signal controller (Starfish Scanline), global positioning unit and control computer.

In the process of image capture or transmission of side scan sonar detection, noise points will be generated due to pulse interference. However, the quality of the side scan sonar image depends on the number of noise pixels in it. Interference sources such as the noise of the propeller and engine wake of the measuring ship, the audio generated by marine organisms in nature, and the background noise that exists in the marine environment, such as the movement of the crust and the wind and waves on the sea surface. Such noises will cause interference to the transmission of sound waves during the transmission of the side scan sonar signal, and these interferences will appear as noise in the side scan sonar image. Therefore, how to filter the noise of the side scan sonar image is one of the very important issues to be solved by the present invention.

In addition, the seabed edge detection provided by the side scan sonar system is used to determine the first return water wave and estimate the seabed edge. However, the first echo is easily affected by suspended solids, wake bubbles, and seabed topography, causing nearly 90% of seabed edges to be marked incorrectly. Therefore, to propose a new seabed edge detection technology to improve the misjudgment of the distance between the towfish and the seabed by the first echo is another issue that the present invention has to face.

Image processing of side scan sonar images for noise and first echo interference will help the subsequent research on seabed edge detection technology. The seabed edge detection technology is used to mark the seabed edge. In addition to providing the correct distance information between the towing fish and the seabed, it can ensure that the towing fish will not touch the seabed, and it can also increase the image quality of the side scan sonar ( The quality of the sonar measured when the fish is dragged close to the seabed is better).

The existing side scan sonar system cannot filter the black horizontal noise caused by ship noise interference, and there is still a lot of room for improvement in the seabed edge detection technology. It is a problem that all parties want to solve and urgently need to be improved. .

In order to solve the aforementioned problems, the object of the present invention is to provide a technical solution for detecting the edge of the seabed that can filter the horizontal noise.

To achieve the above objective, the present invention proposes a seabed edge detection method that can filter horizontal noise, which includes the following steps: (a) the step of inputting grayscale images, when the side scan sonar control computer receives the seabed echo Then, the echo decibel value is converted into a side scan sonar image (gray-scale image). (b) Image pre-processing step, including: (b1) Image sharpening step, the filter kernel passes through the original pixels, and the filter kernel size ( kernal size, ksize ) is k * k filter kernel pixel values and the grayscale image The original pixel values are multiplied in order, and k * k product pixel values are obtained after the multiplication, and then summed in order. If the sum is greater than the pixel value 255, the sum is regarded as the upper limit 255; and (b2) the image blurring step, the sharpened pixel values are processed by a smoothing filter (burring) to produce a blurred side scan sonar image; and finally through the (b3) Otsu Threshold step , To segment the blurred side scan sonar image. In (c) the seabed edge detection and marking step, the Otsu binarized side scan sonar image uses a sliding window method to slide in port and starboard directions to detect the Otsu binarized side scan The pixel value changes in the sonar image to detect the edge features of the seabed.

The advantageous effect that the present invention can achieve is to filter out noise through and (b) the image pre-processing step to enhance the seabed edge line characteristics; and (c) the seabed edge line detection and marking step, using a sliding window method Detect the location of the seabed edge and mark it. Through the detection and marking of the seabed edge of the present invention, a side scan sonar image marked with the seabed edge is obtained, which assists in the detection of underwater terrain and landforms, and more accurately calculates the distance between the tow fish and the seabed terrain .

S10-S40: steps

10: Drag fish

11: Port side sonar signal

12: Starboard sonar signal

20: Port side

30: starboard

40: Noise

50: Seabed

51: Port Seabed

52: Starboard Seabed

60: Port side water mass

70: Starboard Water Mass

80: height

90: target

100: shadow

110: Shipwreck

120: Port side seabed edge

130: Starboard Seabed Edge

140: Seabed Objects

150: foreground

151: Port Side Binary Seabed

152: Starboard Binary Seabed

160: background

161: Port Side Binary Water Group

162: Starboard Binary Water Group

170: Port side sliding window

180: Starboard sliding window

190: Column

200: column

A, B: horizontal noise

[Figure 1] is a block flow diagram of the present invention.

[Figure 2] is a schematic diagram of noise interference in underwater environments or ship engines.

[Figure 3] is a schematic diagram of seabed and water mass imaging.

[Figure 4] is a schematic diagram of sinking ship and noise interference.

[Figure 5] is a comparison chart before and after image sharpening.

[Figure 6] Comparison of the results of low-pass filtering with or without image sharpening.

[Figure 7] is a schematic diagram of Otsu binarization processing.

[Figure 8] The result of Otsu binarization with or without image sharpening and low-pass filtering.

[Figure 9] is a schematic diagram of the seabed edge determination.

[Figure 10] is the result of detection and marking of the seabed edge.

[Figure 11] is a graph showing the results of continuity rate and symmetry rate.

[Figure 12] is the average of continuity rate and symmetry rate.

Specific embodiments will be described below to illustrate the implementation of the present invention, but they are not used to limit the scope of protection of the present invention.

definition

Sonar signal: Please refer to Figure 2. The towfish 10 emits sonar signals from the port side of the stern 20 to lower left and the starboard side of the stern 30 to lower right. The sonar signal emitted by the towfish 10 from the port side of the stern is the port side. The sonar signal 11, the sonar signal emitted by the towfish 10 from the starboard side 30 of the stern is the starboard sonar signal 12. The sonar signal is transmitted through the water, and after reaching the seabed 50, the sonar signal is reflected back to the towfish 10.

Seabed: Please refer to Figure 3. Towfish 10 is the core detection tool of the entire side scan sonar system. Towfish 10 uses port and starboard side drums to emit sonar signals to reach the seabed 50; then, Towfish 10 After the device receives the reflected sonar signal, the signal controller transmits the reflected sonar signal to the control computer after the analog and digital signal conversion, and presents the detected underwater terrain and landforms from top to bottom in a waterfall style, that is, side scan sonar The presentation method of grayscale images. In more detail (please refer to Figure 3), the port side sonar signal 11 is transmitted through the water to the seabed 50 and then reflected back to the towfish 10. The port side sonar signal 11 is echoed to the seabed 50 through the signal controller. The energy value (dB) Converted into visualized port side seabed 51 information; starboard sonar signal 12 is transmitted through the water to the seabed 50 and then reflected back to towfish 10, and all seabed echo energy values (dB) of starboard sonar signal 12 are converted by the signal controller 52 information on the starboard side seabed into a visualization.

Water mass: Please refer to Figure 3. The bottom part of Figure 3 is part of the original side scan sonar image. After the first beam is launched, the length of time that it passes to the seabed 50 in the water and is reflected back to the towfish 10 is used to calculate the distance 80 between the towfish 10 and the seabed 50, which is called a water mass. In more detail (please refer to Figure 4a), the water group includes a port side water group 60 and a starboard side water group 70. The port side water group 60 and the starboard side water group 70 are located between the port side sea bed 51 and the starboard side sea bed 52.

Port side seabed sideline: Please refer to Figure 4a, the port side seabed sideline 120 is located between the port side seabed 51 and the port side water mass 60, which is called the port side seabed sideline 120.

Starboard seabed edge: Please refer to Figure 4a. The starboard seabed edge 130 is located between the starboard seabed 52 and the starboard water mass 70, and is called the starboard seabed edge 130.

Horizontal noise: Please refer to the red frame in Figure 4b, which is the original side scan sonar image. Among them, the horizontal noise A is distributed on the port seabed 51, the port side seabed edge 120 and the port side water mass 60; the horizontal noise B is distributed on the starboard seabed 52, the starboard seabed edge 130 and the port water mass 70.

The present invention provides a seabed edge detection method capable of filtering horizontal noise. Please refer to FIG. 1 for the method and flow. The method includes the following steps: (a) Input side scan sonar grayscale image step S10, wherein the grayscale image Contains the seabed and water mass information, as well as possible noise interference (such as horizontal noise A and B). The seabed edge is located between the seabed and the water mass, and the horizontal noise is distributed on the seabed and sea. Bedside line and water mass; (b) image pre-processing step S20, including image sharpening S21 to generate a sharpened image (see Figure 5), and the sharpened image is subjected to smoothing filter blur processing S22 (please Refer to Figure 6), filter the horizontal noise A and B and smooth the seabed image to produce a blurred seabed image, and then perform Otus binarization segmentation of the blurred seabed image to generate a segmented image S23, where the segmented image Including the foreground 150 part and the background 160 part (see Figure 7, the foreground 150 part is the water mass, and the background 160 part is the seabed); and (c) the seabed edge detection and marking step S30, using the sliding window ( sliding window) Detect the pixel conversion of the water mass and seabed of the segmented image. If the number of pixel conversion matches is once, it is determined that the sea is found. The edge of the bed and mark the edge of the sea bed, and then obtain the finished marking the edge of the sea bed.

Among them, in step S10 of (a) inputting the grayscale image, when the side scan sonar control computer receives the seabed echo signal strength (dB), it will be converted into a side scan sonar image according to the signal strength, where the control computer It can present side scan sonar images with different single tones. In this step S10, the control computer converts the intensity of the side scan sonar echo signal to the grayscale image, which is beneficial to the subsequent (b) image preprocessing step S20 and (c) seabed edge detection and marking step S30. efficient.

Wherein, in the image sharpening S21 of (b) image preprocessing step S20, the original side scan sonar image is input and a storage space is created for storing the sharpened image after the image sharpening process. The size of the storage space ( pixel * pixel ) is the same as the image size (pixel * pixel ) of the original side scan sonar image. Then initialize the initial position of the filter core (Filter), so that the scanning positions of the filter core all start from the initial position (0, 0) of the original side scan sonar image. After that, the filter kernels are slid through and cover each original pixel value in the entire original side scan sonar image, and the filter kernel size ( kernal size, ksize ) k * k filter kernel pixel values are compared with the original side scan sonar image The original pixel values of the image are multiplied in sequence, and k * k product pixel values are obtained after the multiplication. The sum is sequentially summed. If the sum is greater than the pixel value 255, the sum is regarded as the upper limit The value is 255. The original pixel value is processed by image sharpening to obtain the sharpened pixel value. The calculation method is as shown in equations (1) to (2):

方程式(1)中Filter _ksize 為可任意變動之濾波核大小，其中s為用戶自定義值，方程式(2)以濾波核大小3*3為例，src(x,y)為該原始像素點(x,y)處的原始像素值；dst(x,y)為該原始像素點src(x,y)經影像銳化計算後之銳化像素值；Filter為進行遮罩計算的濾波核，其尺寸可任意變動。原始像素值經由影像銳化計算後產生銳化像素值，銳化像素點構成經影像銳化的側掃聲納影像。

Filter _ksize in equation (1) is the filter kernel size that can be changed arbitrarily, where s is a user-defined value, equation (2) takes the filter kernel size 3*3 as an example, and src ( x, y ) is the original pixel ( The original pixel value at x, y ); dst ( x, y ) is the sharpened pixel value of the original pixel src ( x, y ) after the image sharpening calculation; Filter is the filter kernel for mask calculation, which The size can be changed arbitrarily. The original pixel values are calculated by image sharpening to generate sharpened pixel values, and the sharpened pixels constitute a sharpened side scan sonar image.

Further, the filter kernel size ( kernal size, ksize ) used in the present invention is 5*5, and the method of gradually increasing to the center of the filter kernel (Anchor) is adopted. The value design of the filter kernel is as follows, and the filter kernel The value within is designed to satisfy equation (3):

K為用戶自定義值，其範圍約莫在1-5內，皆能處理80%以上的側掃聲納影像。在本發明的較佳實施例中，使用濾波核大小為5*5進行影像銳化，側掃聲納影像銳化結果如圖5所示，圖5a為影像進行銳化前的結果，圖5b為影像進行銳化後的結果。其中，影像進行銳化後的結果明顯可見邊緣輪 廓增強的效果；再者，使用影像銳化，使得海床像素與水團像素之間的邊線越來越明顯，達到側掃聲納影像的海床邊線加強之效果。

K is a user-defined value, and its range is about 1-5, and it can process more than 80% of the side scan sonar images. In a preferred embodiment of the present invention, a filter kernel size of 5*5 is used for image sharpening, and the result of side scan sonar image sharpening is shown in Figure 5. Figure 5a is the result before image sharpening, Figure 5b The result of sharpening the image. Among them, the result of image sharpening clearly shows the effect of edge contour enhancement; Moreover, the use of image sharpening makes the edge line between the seabed pixels and the water mass pixels more and more obvious, achieving the sea side scan sonar image. The effect of strengthening the bedside line.

In the image blurring S22 of (b) image preprocessing step S20, the present invention uses a low-pass filter to filter out noise and smooth the image. The object of noise filtering is mainly the first echo interference and horizontal noise. Please refer to the red box in Figure 4a for the first echo interference. The sunken ship 110 on the starboard seabed edge 130 makes the starboard seabed edge 130 produce pixels similar to 52 pixels on the starboard seabed, thereby affecting the seabed The correctness of the edge detection; please refer to the red box in Figure 4b, where the seabed object 140 in the starboard seabed 52 is disturbed by the noise 40, causing the seabed object 140 to be covered by the horizontal noise B, so In the process of identifying the underwater seabed object 140, the accuracy rate will be greatly reduced. In addition, the pixel values of the horizontal noise A and B are similar to the pixel values of the water mass. Therefore, during the seabed edge detection process, the accuracy of the seabed edge detection will also be affected.

The calculation method using low-pass filter is shown in equations (4) to (5):

其中，src(x,y)為銳化像素點(x,y)處的銳化像素值；dst(x,y)為該銳化像素點(x,y)經低通濾波器計算後之平滑像素值；Filter為進行遮罩計算的濾波核，其尺寸可任意變動；FilterWidthSize為濾波核寬度，FilterHeightSize為濾波核高度，濾波核寬與高指的是濾波核的大小。銳化像素值經由低通濾波器計算用以產生平滑像素值，平滑像素點構成側掃聲納模糊影像。在本發明的演算法實施例中，使用濾波核大小為 17*17進行影像平滑，側掃聲納影像平滑結果如圖6所示，圖6a為無做影像銳化之低通濾波處理結果，圖6b為有做影像銳化之低通濾波處理結果。由圖6b明顯可見，先經影像銳化處理後，再進行低通濾波處理，可增加海床與水團之間的像素值對比度，進而達到濾除橫紋雜訊與平滑海床之效果。

Wherein, sharpened pixel src (x, y) is sharpened pixel (x, y) at a value; dst (x, y) for sharpening the pixel (x, y) is calculated by the low-pass filter Smooth pixel value; Filter is the filter core for mask calculation, and its size can be changed arbitrarily; FilterWidthSize is the width of the filter core, FilterHeightSize is the height of the filter core, and the width and height of the filter core refer to the size of the filter core. The sharpened pixel value is calculated by a low-pass filter to generate a smooth pixel value, and the smooth pixel points form a side-scan sonar blurred image. In the algorithm embodiment of the present invention, a filter kernel size of 17*17 is used for image smoothing. The result of side scan sonar image smoothing is shown in Fig. 6, and Fig. 6a is the result of low-pass filtering without image sharpening. Figure 6b shows the result of low-pass filtering with image sharpening. It is obvious from Fig. 6b that the image sharpening process and then the low-pass filtering process can increase the pixel value contrast between the seabed and the water mass, thereby achieving the effect of filtering horizontal noise and smoothing the seabed.

In (b) the Otsu binarization S23 (Otsu Threshold) of the image preprocessing step S20, the smoothed side scan sonar image is image segmented. Otsu Binarization S23 includes the following steps: (b1) According to the pixel value of the smoothed side scan sonar image, calculate the histogram of the pixel value, wherein the smooth pixel points formed by the smoothed pixel values are calculated according to statistics Method, distributed in a total of 256 levels from 0 to 255 (pixel), as shown in Figure 7; (b2) Normalize the number of pixels in each pixel level, and compress the number of pixels to 0- The interval of 1, in order to facilitate the quantification of the number of pixels in each pixel level calculated; (b3) According to all the smoothed pixel values between pixel levels 1 to 254 and the number of smoothed pixels after quantization, The calculation equations (6) to (11) of Otsu binarization are used to find the optimal threshold T. Among them, all the smoothed pixel values between pixel levels 1 to 254 are used as the threshold T ^* , and the intra-cluster variation is calculated for each threshold T ^* , and then the minimum intra-cluster variation is found and judged as Optimal threshold T (shown by the red line in Figure 7); (b4) According to the optimal threshold T , the smoothed pixel values in the histogram are scanned and scanned sonar image segmentation, where if the smoothed pixel value is greater than or equal to the optimal threshold T , The smooth pixel value is judged as the foreground, otherwise, the smooth pixel value is judged as the background; (b5) In the side scan sonar image obtained by Otsu binarized image segmentation, the foreground is the seabed (the pixel value is 255), and the background It is a water mass (the pixel value is 0). The calculation method of Otsu binarization is shown in equations (6) to (11):

σ( T ^* )=σ ₁ ( T ^* )+σ ₂ ( T ^* ) equation (10)

其中，h(i)代表直方圖第i個像素級別之歸一化的平滑像素點數量；方程式(6)中u ₁(T ^*)代表當T ^*為閾值，計算平滑像素值從0至T ^*之平均值；方程式(7)中u ₂(T ^*)則代表當T ^*為閾值，計算平滑像素值從T ^*至254之平均值；方程式(8)中σ₁(T ^*)則代表當T ^*為閾值，計算平滑像素值從0至T ^*之變異量；方程式(9)中σ₂(T ^*)則代表計算平滑像素值從T ^*至254之變異量；方程式(10)中σ(T ^*)則代表當T ^*為閾值，計算其群內變異數之總和；方程式(11)則代表從0-254的閾值中找到其最小群內變異數。

Among them, h ( i ) represents the number of normalized smooth pixels at the i- th pixel level of the histogram; u ₁ ( T ^* ) in equation (6) represents when T ^* is the threshold, calculate the smooth pixel value from 0 to T ^{The average value of *} ; u ₂ ( T ^* ) in equation (7) means that when T ^* is the threshold, calculate the average of smoothed pixel values from T ^* to 254; in equation (8), σ ₁ ( T ^* ) means When T ^* is the threshold, calculate the variation of the smoothed pixel value from 0 to T ^* ; in equation (9) σ ₂ ( T ^* ) represents the calculated variation of the smoothed pixel value from T ^* to 254; in equation (10) σ( T ^* ) means that when T ^* is the threshold, calculate the sum of the variance within the cluster; Equation (11) means finding the minimum variance within the cluster from the threshold of 0-254.

The Otsu binarization result is shown in Fig. 8. Fig. 8a is the result without image sharpening and low-pass filtering, and Fig. 8b is the result after image sharpening and low-pass filtering. It can be seen from Figure 8a that if the noise filtering process of image sharpening S21 and image blurring S22 is not performed first, the image segmentation effect of the Otsu binarization will be affected by the horizontal noise A and B generated by the ship noise 40. Causes image segmentation errors; relatively, it is obvious from Figure 8b that the horizontal interference caused by the interference of ship noise 40 Pattern noise A and B, through the image sharpening S21 and image blur S22 noise filtering process, the port side seabed 151 and the port side water mass 161 are binarized, and the starboard seabed is binarized The image segmentation between 152 and the binarized starboard water mass 162 helps to improve the accuracy of image segmentation and facilitates the subsequent detection and marking of the seabed edge.

，右舷滑動視窗180正向滑動順序為C _n →

，C為欄200(c3)S32L、S32R和S33之流程，包含以下步驟：(c3.1)當左舷滑動視窗170或右舷滑動視窗180偵測到滑動視窗涵蓋範圍內，發生前景150(像素值為255)轉換成背景160(像素值為0)變化時，且發生像素值變化的次數符合為一次，則該左舷滑動視窗170或右舷滑動視窗180將會停止，並判斷當下左舷滑動視窗170或右舷滑動視窗180之滑動視窗涵蓋範圍，已遮蓋在左舷海床邊線120或右舷海床邊線130之上；反之，若像素值發生變化的次數大於一次，則左舷滑動視窗170或右舷滑動視窗180會繼續往前偵測，直到符合像素值發生變化的次數為一次；(c3.2)滿足符合像素值發生變化的次數為一次的條件後，接著左舷滑動視窗170或右舷滑動視窗180進行反向偵測，反向偵測的滑動順序與正向偵測相反。其中，反向偵測設定之滑動視窗大小(window_size)小於正向偵測設定之滑動視窗大小，反向偵測的設定步伐為1個像素值之大小(window_size=1)；(c3.3)在步驟 S33，當左舷滑動視窗170或右舷滑動視窗180偵測到滑動視窗涵蓋範圍內發生背景160轉換成前景150的像素值變化時，則將該像素值變化之所在像素視為海床邊線；(c3.4)在步驟S34，將滑動視窗偵測到的發生像素值變化處進行標記。其中，標記海床邊線的滑動順序為R _i →R _n ，R為列190；(c3.5)在步驟S35判斷是否為最後一列R _n ，由於影像輸入後，首先會去讀取該灰階影像的尺寸大小，進而取得其總列數R _n 。當判斷為最後一列R _n 且完成R _n 的海床邊線標記後，則停止標記；(c3.6)在步驟S36，獲得左舷海床邊線120及右舷海床邊線130。其中，由於每筆海床邊線應屬連續，因此所標記之第i筆海床邊線與i+1筆海床邊線並不會落差太大，在影像呈現上形成一條連續的左舷海床邊線120與一條連續的右舷海床邊線130，如圖10所示。 In (c) Seabed edge detection and marking step S30, it includes the following steps: (c1) Binarize the side scan sonar image based on Otsu and use the sliding window to find seabed edge features S31 , Please refer to Figure 9. Among them, the Otsu binarized side scan sonar image is divided into two parts on the left and right of the image center axis, respectively detecting the port side seabed edge line 120 and the starboard side seabed edge line 130; (c2) S32L and S32R use the port side respectively Sliding window and starboard sliding window, from Otsu binarized side scan sonar image, forward detection to the image center axis on both sides of the image center axis to find the starboard seabed sideline 130 and the port side seabed Sideline 120. The forward sliding sequence of the port side sliding window 170 is C _i →

, The forward sliding sequence of the starboard sliding window 180 is C _n →

, C is the flow of column 200 (c3) S32L, S32R and S33, including the following steps: (c3.1) When the port side sliding window 170 or starboard side sliding window 180 detects the sliding window coverage, the foreground 150 (pixel value) 255) is converted to background 160 (pixel value 0) changes, and the number of pixel value changes coincides with once, then the port side sliding window 170 or starboard side sliding window 180 will stop, and the current port side sliding window 170 or The sliding window coverage of the starboard sliding window 180 has been covered on the port side seabed sideline 120 or the starboard side seabed sideline 130; on the contrary, if the pixel value changes more than once, the port side sliding window 170 or the starboard side sliding window 180 will continue to detect until the number of times the pixel value has changed is once; (c3.2) After meeting the condition that the number of times the pixel value has changed is once, then the port side sliding window 170 or the starboard side sliding window 180 will be reversed. The sliding sequence of forward detection and reverse detection is opposite to that of forward detection. Wherein the sliding window size is set to detect the reverse of (window _size) of less than the forward sliding detection window size is set, the pace is set reverse to detect a size of the pixel values _{(window size = 1); (} c3. 3) In step S33, when the port side sliding window 170 or the starboard side sliding window 180 detects that the background 160 is converted into the foreground 150 pixel value within the scope of the sliding window, the pixel where the pixel value changes is regarded as the seabed Borderline; (c3.4) In step S34, mark the pixel value change detected by the sliding window. Among them, the sliding sequence for marking the edge of the seabed is R _i → R _n , and R is column 190; (c3.5) In step S35, it is judged whether it is the last column R _n . After the image is input, the gray will be read first. The size of the first-level image, and then obtain the total number of rows R _n . When _n is determined, and the last a complete R R _n line marking the sea bed, the stop flag; (C3.6) In step S36, obtaining the port and starboard sea bed sea bed 120 line 130 line. Among them, since each seabed edge should be continuous, the marked i-th seabed edge and the i +1 seabed edge will not be too far apart, forming a continuous port side sea on the image presentation The bedside line 120 and a continuous starboard side seabed line 130 are shown in FIG. 10.

window _size = aSize * kSize equation (12) where window _size is the sliding window; aSize and kSize are the sizes of the sliding window that can be changed arbitrarily.

其中window _i 為滑動至第i步的滑動視窗，k為化動視窗滑動步伐大小，window(x,y)為滑動視窗內第一個像素值，window(x _j ,y)為滑動視窗內最後一個像素值，當像素值相減絕對值等於0，就視為滑動視窗尚未遮蓋於海床邊線上，若像素值相減絕對值等於255，就視為滑動視窗已遮蓋於海床邊線上。

Among them, window _i is the sliding window sliding to the i-th step, k is the size of the sliding step of the moving window, window ( x, y ) is the first pixel value in the sliding window, window ( x _j , y ) is the last in the sliding window For a pixel value, when the absolute value subtracted from the pixel value is equal to 0, it is deemed that the sliding window has not been covered on the seabed edge. If the absolute value subtracted from the pixel value is equal to 255, it is deemed that the sliding window has been covered on the seabed edge.

當先前條件成立，則從window(x _j ,y)繼續往後看n個像素，n為用戶自定義值。若往後看n個像素之像素值皆為0，則視為該滑動視窗已遮該於海床邊線上且變換次數為一次，而往後看n個像素之像素值有出現255，則反之。

When the previous condition is established, continue to look at n pixels from window ( x _j , y ), where n is a user-defined value. If the pixel values of n pixels are all 0 in the future, it is deemed that the sliding window has been hidden on the edge of the seabed and the number of transformations is once, and the pixel values of n pixels in the future have 255, and vice versa .

接著會對該滑動視窗之window(x _j ,y)座標位子進行反向偵測，當window(x _j -1,y)為0則視該像素值非為海床邊線點，若window(x _j -1,y)為255則視該像素為海床邊線點。

Then the window ( x _j , y ) coordinate position of the sliding window will be detected in the reverse direction. When window ( x _j -1 , y ) is 0, the pixel value is regarded as a seabed edge point. If window ( If x _j -1 ,y ) is 255, the pixel is regarded as a seabed edge point.

It is supplemented that the seabed edge obtained by the present invention is verified by calculating the continuity rate and the symmetry rate (non-patent literature: Z. Jianhu, W. Xiao, Z. Hongmei, and W. Aixue, A Comprehensive Bottom -Tracking Method for Side Scan Sonar Image Influenced by Complicated Measuring Environment, IEEE Journal of Ocenic Engineeing , November 2016). Among them, there are a total of 130 side scan sonar grayscale images analyzed. Figure 11a shows the 120 continuity rate of the port side seabed edge, with an average effect of 87.8%; Figure 11b shows the 130 continuity rate of the starboard seabed edge, with an average of 94.8%. The effect; Figure 11c shows the symmetry rate of the seabed edge, with an average effect of 80.7%. Figure 12 shows the average value of the continuity rate of the port side seabed edge 120 and the starboard side seabed edge 130 of the 130 gray-scale images, and the average value of the seabed edge symmetry rate.

The above detailed description is a specific description of a possible embodiment of the present invention, but this embodiment is not intended to limit the scope of the present invention. Any equivalent implementation or modification that does not deviate from the spirit of the method of the present invention should be included in In the scope of the patent in this case.

S10-S40: steps

Claims (8)

A seabed edge detection method that can filter horizontal noise. It includes: (a) Input grayscale image step, when the side scan sonar control computer receives the seabed echo, it will convert it into the original side scan sonar image (Gray-scale image); (b) Image pre-processing step, including: image sharpening step, where the pixel value of the original pixel is used to generate sharpened pixel value through image sharpening calculation, where the image sharpening calculation includes filtering The kernel covers the original pixels in the gray-scale image, and sequentially multiplies the filter kernel size (kernal size, ksize ) k * k filter kernel pixel values with the original pixel values of the gray-scale image; After the operation, k * k product pixel values are obtained in order for the sum operation; the image blurring step, the sharpened pixel values are processed by the filter to generate complex smooth pixel values, and these smooth pixel values constitute a side sweep sound Nano-blur image; Otsu threshold (Otsu Threshold) step, the side scan sonar blurred image is segmented through Otsu binarization to generate an Otsu binary side-scan sonar image; and (c) The seabed edge detection and marking steps are based on the Otsu binarized side-scan sonar image and used to find the seabed edge features through a plurality of sliding windows, wherein the sliding windows are used to detect the The change of pixel values in the Otsu binarization side scan sonar image; among them, the Otsu binarization step of (b) image preprocessing before image segmentation further includes the following steps: according to the smoothed pixel values and the number of pixels It is used to calculate its histogram (Histogram), in which the complex smooth pixel points formed by the smooth pixel values are distributed in a statistical manner from 0 to 255 (pixel) levels; the number of pixels in each pixel level is returned One conversion; according to all the smoothed pixel values between pixel level 1 to 254 and the number of smoothed pixels after quantization, the calculation equation of Otsu binarization is used to find the best threshold T; according to the best threshold T to the histogram The smoothed pixel values of the image segmentation are performed. The seabed edge detection method described in claim 1, wherein the equation used in the image sharpening step of (b) image preprocessing is as follows:

The seabed edge detection method described in claim 1, wherein the equation used in the image blurring step of (b) image preprocessing is as follows:

The seabed edge detection method according to claim 1, wherein the size of the filter kernel used in the image sharpening step is 5*5. The seabed edge detection method according to claim 1, wherein the size of the filter kernel used in the image sharpening and blurring step is 17*17. The seabed edge detection method according to claim 1, wherein the sliding windows for detecting changes in pixel values in the Otsu binarized side scan sonar image further include the following steps: the sliding The window is detected in a positive direction from the two sides of the Otsu binarized side scan sonar image relative to the image center axis toward the image center axis; when the sliding windows detect that the pixel value changes within the scope of the sliding window, And the number of pixel value changes coincides with one time, the sliding windows will stop sliding; if the number of pixel value changes is greater than one, the sliding windows continue to slide until the number of pixel value changes coincides with one time . The seabed edge detection method described in claim 6, wherein, after the condition that the number of pixel value changes is one time is met, the method further includes the following steps: then the sliding windows perform reverse detection, and The window _{size of the} forward detection setting is smaller than the sliding window size of the forward detection setting; when the sliding windows detect that the pixel value changes within the scope of the sliding window, and the number of pixel value changes accords with Once, the sliding windows will stop sliding and mark them. Such as the seabed edge detection method described in claim 1, 6 or 7, wherein the calculation equations for the size of the sliding windows are as follows, where the values of aSize and kSize can be defined by the _{size of the} sonar image: window size = aSize * kSize .

Images