CN106296603B

CN106296603B - A kind of small target detecting method of underwater sonar image

Info

Publication number: CN106296603B
Application number: CN201610629838.6A
Authority: CN
Inventors: 郑丽颖; 田凯
Original assignee: Harbin Engineering University
Current assignee: Harbin Engineering University
Priority date: 2016-08-03
Filing date: 2016-08-03
Publication date: 2019-03-05
Anticipated expiration: 2036-08-03
Also published as: CN106296603A

Abstract

The invention belongs to object detection fields, and in particular to a kind of small target detecting method of underwater sonar image.The present invention includes: initialization: enabling binarization threshold th=0.8, shade direction is θ degree, enables I₀=I；Enable SE₁Indicate target template, SE₁It is to enable SE having a size of the square templates of 3 × 3 pixels₀Indicate background template, SE₀It is the fringe region of the square templates having a size of 25 × 25 pixels.Target detection template proposed by the invention is simple；It is fast to detect speed；Noise resisting ability is strong；Without training data.

Description

A kind of small target detecting method of underwater sonar image

Technical field

The invention belongs to object detection fields, and in particular to a kind of small target detecting method of underwater sonar image.

Background technique

Underwater sonar system should also have efficient Target detection and identification energy in addition to needing competent extreme operating environments Power can be differentiated accurately and sink to the bottom Small object and water Small Target.However, being due to sonar system received echo at a certain moment The vector sum of underwater each reflected acoustic wave, the echo amplitude of system record are in wobble variation, generate spot accordingly when forming sound spectrogram Noise influences the explanation to sound spectrogram.Although side-scan sonar can provide high resolution graphics as imaging sonar technology reaches its maturity Picture, but compared with optical imagery, side-scanning sonar image still have between poor image quality, target and background contrast low etc. Problem.Especially for underwater small target deteection problem, due to the influence of interchannel noise and shot noise, so that existing target Detection technique is extremely difficult to satisfied result.

Small object is detected from underwater sonar data using image processing techniques and mode identification technology from last century 90 years Rapid progress has been achieved since generation.These methods can be generally divided into two classes: have measure of supervision and unsupervised approaches.Wherein have Measure of supervision needs a large amount of training datas, and algorithm performance height relies on the property of training data.Such methods mainly include K- close Adjacent neural network, best identification filtered classification device, Bayes classifier, principal component analysis etc..On the other hand, unsupervised approaches without Special training data is needed, it is data-optimized not for special training suitable for the input data of suitable wide scope.Such method Including technologies such as Markov random field, statistics snake model, Morphological scale-spaces.In addition, in order to further increase algorithm performance, There is scholar to blend a variety of different detection/recognition technologies, the integration technology mainly used includes: by the inspection of various algorithms Survey/recognition result is merged using logic fusion (such as boolean with or operation), log-likelihood ratio testing algorithm, linear weighted function.

The present invention utilizes the morphological transformation with stronger anti-noise ability --- and occupation rate hits-miss transformation (Percentage Occupancy Hit-or-Miss Transform, POHMT) realizes that sound is swept in a kind of fast and accurately side Receive image small target deteection algorithm.

Summary of the invention

The purpose of the present invention is to provide one kind to hit-miss transformation (Percentage based on occupation rate Occupancy Hit-or-Miss Transform, POHMT) underwater sonar image small target detecting method.

The object of the present invention is achieved like this:

(1) it initializes: enabling binarization threshold th=0.8, shade direction is θ degree, enables I₀=I, wherein I is H × W pixel Input sonar image matrix；

(2) SE is enabled₁Indicate target template, SE₁It is to enable SE having a size of the square templates of 3 × 3 pixels₀Indicate background mould Plate, SE₀It is the fringe region of the square templates having a size of 25 × 25 pixels；

(3) image I is calculated₀Luminance mean value, use m₀It indicates, enables 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 are calculated separately

PoFore (t)=conv (B_t,SE₁)/9

PoBack (t)=conv (1-B_t,SE₀)/96

Wherein, t=m₀... 255, conv () indicate convolution algorithm, for fixed t value, poFore (t) and poBack It (t) is H × W matrix；

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

Wherein, min { } is minimum value function, and min { poFore (t), poBack (t) } indicates the t fixed for one Value takes among preceding poFore (t) and poBack (t) the two lesser one；Max { } is max function；PoHMT is H × W Matrix；

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

Wherein, h=1 ... H, w=1 ... W, poHMT (h, w) indicate value of the two-dimensional matrix poHMT at (h, w), poBW For two values matrix, poBW (h, w) indicates value of the poBW at (h, w)；

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

(8) poBW is enabled₁It indicates the poBW after denoising, marks poBW using 8- join domain labeling method₁In connected region Domain records position, the length and width of the boundary rectangle of each connected region respectively；

(9) I is enabled₀=255-I is executed step (3)~step (7), and is used poBW₀It indicates the poBW after denoising, utilizes 8- Join domain labeling method label label poBW₀In connected region, record respectively the boundary rectangle frame of each connection position, Length and width；

(10) with poBW₁In a connected region R₁Boundary rectangle be foundation, determine region of search；This is searched for Area maps are to poBW₀Corresponding region, and indicate poBW with RS₀In this region of search；

(11) in poBW₀The region RS in scan for, if it is possible to find a connected region R₀, then by poBW₁In Connected region R₁Labeled as target, while deleting region R₀；Otherwise, if in poBW₀RS in be not present connected region, then will Connected region R in poBW1₁Labeled as non-targeted.

(12) step (10)~step (11) are repeated, checks poBW₁In next connected region, until traversal poBW₁'s All connected regions.

The beneficial effects of the present invention are: 1) the target detection template proposed is simple；2) detection speed is fast；3) antinoise Ability is strong；4) it is not necessarily to training data.

Detailed description of the invention

Fig. 1 is target and its shade schematic diagram in side-scanning sonar image；

Fig. 2 is target template schematic diagram；

Fig. 3 is background template schematic diagram；

Fig. 4 is region of search schematic diagram.

Specific embodiment

The present invention is described further with reference to the accompanying drawing.

The present invention propose one the technical solution adopted is that: 1) initialize: setting detection threshold value, determine target and background template Size, setting shade direction；2) POHMT operation is carried out to former sonar image, and operation result is done into binary conversion treatment, obtain mesh Mark bianry image；3) inverse processing is done to former sonar image；4) POHMT operation is done to the sonar image after inverse, and will fortune It calculates result and does binary conversion treatment, obtain shade bianry image；5) target bianry image, shade bianry image and shade side are utilized To marking Small object.

If I is input sonar image matrix, having a size of H × W pixel, I (h, w) indicates the brightness of image at the position (h, w), Wherein h=1,2 ..., H, w=1,2 ..., W；It and include multiple high luminance targets regions in image I, each target all has Corresponding low-light level shadow region, as shown in Figure 1.

The specific implementation steps are as follows for object detection method proposed by the invention:

Step 1, it initializes: enabling binarization threshold th=0.8, shade direction is θ degree, enables I₀=I.

Step 2, SE is enabled₁Indicate target template, here SE₁It is the square templates having a size of 3 × 3 pixels, as shown in Figure 2； Enable SE₀Indicate background template, here SE₀It is the fringe region of the square templates having a size of 25 × 25 pixels, as shown in figure 3, figure In 3, the size of gray area is 23 × 23 pixels, and pixel value 0, the pixel value of edge white area is 1.

Step 3, image I is calculated₀Luminance mean value, use m₀It indicates；Enable t=m₀... 255, for each t, according to formula (1) bianry image B is calculated_t:

Wherein, h=1 ... H, w=1 ... W.

Step 4, foreground template matching rate poFore is calculated using formula (2), calculates background template using formula (3) and matches Rate poBack:

PoFore (t)=conv (B_t,SE₁)/9 (2)

PoBack (t)=conv (1-B_t,SE₀)/96 (3)

Wherein, t=m₀... 255, conv () indicate convolution algorithm, for fixed t value, poFore (t) and poBack It (t) is H × W matrix.

Step 5 ,-miss transformation poHMT is hit using the occupation rate that formula (4) calculate image:

Wherein, min { } is minimum value function, and min { poFore (t), poBack (t) } indicates the t fixed for one Value takes among poFore (t) and poBack (t) the two lesser one；Max { } is max function；PoHMT is H × W square Battle array.

Step 6, the poHMT of step 5 is done into binary conversion treatment using formula (5)

Wherein, h=1 ... H, w=1 ... W, poHMT (h, w) indicate value of the two-dimensional matrix poHMT at (h, w), poBW For two values matrix, poBW (h, w) indicates value of the poBW at (h, w),

Step 7, the tiny noise spot in morphological dilations, erosion operation removal poBW is utilized.

Step 8, poBW is enabled₁It indicates the poBW after denoising, marks poBW using 8- join domain labeling method₁In company Logical region, records position, the length and width of the boundary rectangle of each connected region respectively.

Step 9, I is enabled₀=255-I executes step 3~step 7, and uses poBW₀It indicates the poBW after denoising, utilizes 8- Join domain labeling method label label poBW₀In connected region, record the position of the boundary rectangle frame of each connected region respectively It sets, length and width.

Step 10, with poBW₁In a connected region R₁Boundary rectangle be foundation, region of search is determined, such as Fig. 4 institute Show, wherein w=5, l=5 × w₀, w₀For connected region R₁Boundary rectangle width.This region of search is mapped to poBW₀Phase Region is answered, and indicates poBW with RS₀In this region of search.

Step 11, in poBW₀The region RS in scan for, if it is possible to find a connected region (assuming that using R₀It indicates This connected region), then by poBW₁In connected region R₁Labeled as target, while deleting region R₀；Otherwise, if poBW₀RS in be not present connected region, then by the connected region R in poBW1₁Labeled as non-targeted.

Step 12, step 10~step 11 is repeated, checks poBW₁In next connected region, until traversal poBW₁'s All connected regions.

Claims

1. a kind of small target detecting method of underwater sonar image, which comprises the steps of:

(1) it initializes: enabling binarization threshold th=0.8, shade direction is θ degree, enables I₀=I, wherein I is the input of H × W pixel Sonar image matrix；

(2) SE is enabled₁Indicate target template, SE₁It is the square templates having a size of 3 × 3 pixels；Enable SE₀Indicate background template, SE₀ It is the fringe region of the square templates having a size of 25 × 25 pixels；

(3) image I is calculated₀Luminance mean value, use m₀It indicates；Enable 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 are calculated separately:

PoFore (t)=conv (B_t,SE₁)/9

PoBack (t)=conv (1-B_t,SE₀)/96

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

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

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

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

Wherein, h=1 ... H, w=1 ... W, poHMT (h, w) indicate value of the two-dimensional matrix poHMT at (h, w), poBW (h, w) Indicate value of the poBW at (h, w)；

(8) poBW is enabled₁It indicates the poBW after denoising, marks poBW using 8- join domain labeling method₁In connected region, Position, the length and width of the boundary rectangle of each connected region are recorded respectively；

(9) I is enabled₀=255-I is executed step (3)~step (7), and is used poBW₀It indicates the poBW after denoising, is connected using 8- Zone marker method marks poBW₀In connected region, record position, the length and width of the boundary rectangle of each connected region respectively；

(10) with poBW₁In a connected region R₁Boundary rectangle be foundation, determine region of search；By this region of search It is mapped to poBW₀Corresponding region, and indicate poBW with RS₀In this region of search；

(11) in poBW₀The region RS in scan for, if it is possible to find a connected region R₀, then by poBW₁In connection Region R₁Labeled as target, while deleting region R₀；Otherwise, if in poBW₀RS in be not present connected region, then by poBW₁ In connected region R₁Labeled as non-targeted；

(12) step (10)~step (11) are repeated, checks poBW₁In next connected region, until traversal poBW₁It is all Connected region.