CN106296603A - A kind of small target detecting method of underwater sonar image - Google Patents

Abstract

The invention belongs to the field of target detection, and in particular relates to a small target detection method for underwater sonar images. The present invention includes: initialization: make the binarization threshold th=0.8, the shadow direction is θ degrees, let I ₀ =I; let SE ₁ represent the target template, SE ₁ is a square template with a size of 3×3 pixels, let SE ₀ Denotes the background template, SE ₀ is the edge region of the square template with dimensions 25 × 25 pixels. The target detection template proposed by the invention is simple; the detection speed is fast; the anti-noise ability is strong; and no training data is needed.

Description

A small target detection method for underwater sonar images

technical field

The invention belongs to the field of target detection, and in particular relates to a small target detection method for underwater sonar images.

Background technique

In addition to being competent in extreme working environments, the underwater sonar system also needs to have efficient target detection and recognition capabilities, and can accurately distinguish between small targets on the bottom and small targets in the water. However, since the echo received by the sonar system at a certain moment is the vector sum of the reflected sound waves underwater, the echo amplitude recorded by the system fluctuates, and speckle noise is correspondingly generated when the acoustic image is formed, which affects the accuracy of the acoustic image. Explanation. Although side-scan sonar can provide high-resolution images with the maturity of imaging sonar technology, compared with optical images, side-scan sonar images still have problems such as poor imaging quality and low contrast between the target and the background. Especially for the underwater small target detection problem, due to the influence of channel noise and scattering noise, it is difficult for the existing target detection technology to achieve satisfactory results.

The detection of small targets from underwater sonar data using image processing techniques and pattern recognition techniques has made great progress since the 1990s. These methods can be broadly classified into two categories: supervised methods and unsupervised methods. Among them, supervised methods require a large amount of training data, and the performance of the algorithm is highly dependent on the nature of the training data. Such methods mainly include K-nearest neighbor neural network, optimal identification filter classifier, Bayesian classifier, principal component analysis and so on. On the other hand, unsupervised methods do not require special training data, are applicable to a fairly wide range of input data, and are not optimized for special training data. Such methods include techniques such as Markov random fields, statistical snake models, and morphological processing. In addition, in order to further improve the performance of the algorithm, some scholars also integrate various detection/recognition technologies. The main fusion technologies include: logical fusion of the detection/recognition results of various algorithms (such as Boolean and, or operations), Log-likelihood ratio testing algorithms, linear weighted fusion, and more.

The present invention utilizes a morphological transformation with strong anti-noise capability——Percentage Occupancy Hit-or-Miss Transform (POHMT) to realize a fast and accurate side-scan sonar image small Object detection algorithm.

Contents of the invention

The object of the present invention is to provide a small target detection method based on an occupancy hit-or-miss transform (Percentage Occupancy Hit-or-Miss Transform, POHMT) underwater sonar image.

The purpose of the present invention is achieved like this:

(1) Initialization: make the binarization threshold th=0.8, the shadow direction is θ degrees, make I ₀ =I, where I is the input sonar image matrix of H×W pixels;

(2) Let SE ₁ represent the target template, SE ₁ is a square template with a size of 3×3 pixels, let SE ₀ represent a background template, and SE ₀ is the edge area of a square template with a size of 25×25 pixels;

(3) Calculate the mean value of the brightness of the image I ₀ , represented by m ₀ , let t=m ₀ ... 255, for each t, calculate the binary image B _t :

Among them, h=1...H, w=1...W;

(4) Calculate the foreground template matching rate poFore and the background template matching rate poBack respectively

poFore(t)=conv(B _t ,SE ₁ )/9

poBack(t)=conv(1-B _t ,SE ₀ )/96

Among them, t=m ₀ ... 255, conv(.) represents a convolution operation, and for a fixed value of t, poFore(t) and poBack(t) are both H×W matrices;

(5) Calculate the occupancy hit-miss transformation poHMT of image I ₀

$\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; 0....255</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \}</span>$

Among them, min{.} is the minimum value function, and min{poFore(t), poBack(t)} means that for a fixed t value, take the smaller one of the former poFore(t) and poBack(t) ;max{.} is the maximum value function; poHMT is an H×W matrix;

(6) Binarize the poHMT of step (5)

Among them, h=1...H, w=1...W, poHMT(h,w) represents the value of the two-dimensional matrix poHMT at (h,w), poBW is a binary matrix, and poBW(h,w) represents poBW The value at (h,w);

(7) Use morphological expansion and erosion operations to remove small noise points in poBW;

(8) Let poBW ₁ represent the poBW after denoising, use the 8-connected region marking method to mark the connected regions in poBW ₁ , and record the position, length and width of the circumscribed rectangles of each connected region respectively;

(9) Set I ₀ =255-I, execute steps (3) to (7), and use poBW ₀ to represent the poBW after denoising, use the 8-connected region marking method to mark the connected regions in poBW ₀ , and record them respectively The position, length and width of each connected circumscribed rectangular frame;

(10) Based on the circumscribed rectangle of a connected region R ₁ in poBW ₁ , determine the search area; map this search area to the corresponding area of poBW ₀ , and use RS to represent the search area in poBW ₀ ;

(11) Search in the RS region of poBW ₀ , if a connected region R ₀ can be found, mark the connected region R ₁ in poBW ₁ as the target, and delete the region R ₀ at the same time; otherwise, if the RS in poBW ₀ There is no connected region in , then the connected region R ₁ in poBW1 is marked as non-target.

(12) Repeat steps (10) to (11) to check the next connected region in poBW ₁ until all connected regions in poBW ₁ are traversed.

The beneficial effects of the invention are: 1) the proposed target detection template is simple; 2) the detection speed is fast; 3) the anti-noise ability is strong; 4) no training data is needed.

Description of drawings

Figure 1 is a schematic diagram of the target and its shadow in the side-scan sonar image;

Fig. 2 is a schematic diagram of a target template;

Figure 3 is a schematic diagram of a background template;

Fig. 4 is a schematic diagram of the search area.

detailed description

The present invention will be further described below in conjunction with the accompanying drawings.

The technical solution proposed by the present invention is: 1) Initialization: setting the detection threshold, determining the size of the target and background template, and setting the shadow direction; 2) performing POHMT calculation on the original sonar image, and binarizing the calculation result to obtain Target binary image; 3) Invert the original sonar image; 4) Perform POHMT operation on the inverted sonar image, and binarize the operation result to obtain a shadow binary image; 5) Use the target Binary image, shadow binary image, and shadow direction to mark small targets.

Let I be the input sonar image matrix, the size is H×W pixels, I(h,w) represents the brightness of the image at the position (h,w), where h=1,2,...,H,w=1,2 ,...,W; and the image I contains multiple high-brightness object regions, each of which has a corresponding low-brightness shadow region, as shown in Figure 1.

The specific implementation steps of the target detection method proposed by the present invention are as follows:

Step 1, initialization: set the binarization threshold th=0.8, the shadow direction is θ degrees, and set I ₀ =I.

Step 2, let SE ₁ represent the target template, where SE ₁ is a square template with a size of 3×3 pixels, as shown in Figure 2; let SE ₀ represent a background template, where SE ₀ is a square template with a size of 25×25 pixels The edge area of , as shown in Figure 3, in Figure 3, the size of the gray area is 23×23 pixels, its pixel value is 0, and the pixel value of the edge white area is 1.

Step 3, calculate the brightness mean value of the image I ₀ , denoted by m ₀ ; let t=m ₀ ... 255, for each t, calculate the binary image B _t according to the formula (1):

Among them, h=1...H, w=1...W.

Step 4, use the formula (2) to calculate the foreground template matching rate poFore, and use the formula (3) to calculate the background template matching rate poBack:

poFore(t)=conv(B _t ,SE ₁ )/9 (2)

poBack(t)=conv(1-B _t ,SE ₀ )/96 (3)

Wherein, t=m ₀ ... 255, conv(.) represents a convolution operation, and for a fixed value of t, both poFore(t) and poBack(t) are H×W matrices.

Step 5, use the formula (4) to calculate the occupancy rate hit-miss transformation poHMT of the image:

$\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; 0....255</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; B</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

Among them, min{.} is the minimum value function, and min{poFore(t), poBack(t)} means that for a fixed t value, take the smaller one of poFore(t) and poBack(t); max{.} is the maximum value function; poHMT is an H×W matrix.

Step 6, use formula (5) to binarize the poHMT in step 5

Among them, h=1...H, w=1...W, poHMT(h,w) represents the value of the two-dimensional matrix poHMT at (h,w), poBW is a binary matrix, and poBW(h,w) represents poBW The value at (h,w),

Step 7, use morphological expansion and erosion operations to remove small noise points in poBW.

Step 8: Let poBW ₁ represent the denoised poBW, use the 8-connected region marking method to mark the connected regions in poBW ₁ , and record the position, length and width of the circumscribed rectangles of each connected region.

Step 9, set I ₀ =255-I, execute steps 3 to 7, and use poBW ₀ to represent the poBW after denoising, use the 8-connected region marking method to mark the connected regions in poBW ₀ , and record the connected regions of each connected region The position, length, and width of the bounding rectangle.

Step 10, based on the circumscribed rectangle of a connected region R ₁ in poBW ₁ , determine the search area, as shown in Figure 4, where w=5, l=5×w ₀ , w ₀ is the circumscribed rectangle of the connected region R ₁ Rectangle width. Map this search area to the corresponding area of poBW ₀ , and denote this search area in poBW ₀ by RS.

Step 11, search in the RS area of poBW ₀ , if a connected area can be found (assuming that this connected area is represented by R ₀ ), then mark the connected area R ₁ in poBW ₁ as the target, and delete the area R ₀ ; Otherwise, if there is no connected region in the RS of poBW ₀ , the connected region R1 in _poBW1 is marked as non-target.

Step 12, repeat steps 10 to 11, check the next connected region in poBW ₁ until all connected regions in poBW ₁ are traversed.

Claims (1)

1. the small target detecting method of a underwater sonar image, it is characterised in that comprise the steps:

(1) initialize: making binary-state threshold th=0.8, shade direction is θ degree, makes I₀=I, wherein I is the input of H × W pixel Sonar image matrix；

(2) SE is made₁Represent To Template, SE₁It it is the square templates of a size of 3 × 3 pixels；Make SE₀Represent background template, SE₀ It it is the marginal area of the square templates of a size of 25 × 25 pixels；

(3) image I is calculated₀Luminance mean value, use m₀Represent；Make t=m₀... 255, for each t, calculate bianry image B_t:

Wherein, h=1 ... H, w=1 ... W；

(4) foreground template matching rate poFore and background template matching rate poBack is calculated respectively:

PoFore (t)=conv (B_t,SE₁)/9

PoBack (t)=conv (1-B_t,SE₀)/96

Wherein, t=m₀... 255, conv (.) represents convolution algorithm；

(5) image I is calculated₀Occupation rate hit-miss conversion poHMT:

$poHMT=\underset{t=m\mathrm{0....255}}{max}\left\{\min \right\{poFore\left(t\right),poBack\left(t\right)\left\}\right\}$

Wherein, min{.} is minimum value function, and max{.} is max function；

(6) poHMT of step (5) is done binary conversion treatment:

Wherein, h=1 ... H, w=1 ... W, poHMT (h, w) represent two-dimensional matrix poHMT (h, w) value at place, poBW (h, w) Represent that poBW is in (h, w) value at place；

(7) morphological dilations, erosion operation is utilized to remove the tiny noise spot in poBW；

(8) poBW is made₁Represent the poBW after denoising, utilize 8-join domain labeling method labelling poBW₁In connected region, Record the position of the boundary rectangle of each connected region, length and width respectively；

(9) I is made₀=255-I, performs step (3)～step (7), and uses poBW₀Represent the poBW after denoising, utilize 8-to connect Zone marker method labelling poBW₀In connected region, record the position of the boundary rectangle of each connected region, length and width respectively；

(10) with poBW₁In connected region R₁Boundary rectangle be foundation, determine region of search,；By this region of search It is mapped to poBW₀Respective regions, and represent poBW with RS₀In this region of search；

(11) at poBW₀RS region in scan for, if it is possible to find connected region R₀, then by poBW₁In connection Region R₁It is labeled as target, deletes region R simultaneously₀；Otherwise, if at poBW₀RS in there is not connected region, then by poBW₁ In connected region R₁It is labeled as non-targeted.

(12) repeat step (10)～step (11), check poBW₁In next connected region, until traversal poBW₁All Connected region.