CN105303526B - A kind of Ship Target Detection method based on coastline data and spectrum analysis - Google Patents
A kind of Ship Target Detection method based on coastline data and spectrum analysis Download PDFInfo
- Publication number
- CN105303526B CN105303526B CN201510594428.8A CN201510594428A CN105303526B CN 105303526 B CN105303526 B CN 105303526B CN 201510594428 A CN201510594428 A CN 201510594428A CN 105303526 B CN105303526 B CN 105303526B
- Authority
- CN
- China
- Prior art keywords
- image
- data
- coastline
- noise
- cloud layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
A kind of Ship Target Detection method based on coastline data and spectrum analysis, the present invention relates to Ship Target Detection method.The problem of larger computing resource of Ship Target in noise and large format satellite image can not effectively be removed the present invention is to solve prior art detection consumption and time, and a kind of Ship Target Detection method based on coastline data and spectrum analysis proposed.This method is to be all the image after the view data of ocean i.e. mask by one, formation background；2nd, obtain removing the image after cloud layer noise；3rd, binary image is obtained, seed point or seed region are detected in binary image；Naval vessel hull areas is obtained according to detecting seed point or seed region and carrying out region growing；4th, screened by the area and aspect ratio features of naval vessel hull areas, obtain what the step such as Ship Target Detection result in final high-resolution satellite image was realized.The present invention is applied to Ship Target Detection field.
Description
Technical field
The present invention relates to Ship Target Detection method, more particularly to a kind of naval vessel based on coastline data and spectrum analysis
Object detection method.
Background technology
With the development of satellite remote sensing technology, obtain high-definition picture and have become one of trend.High-definition picture
Middle target detail is clear, half-tone information enriches, and good data basis is provided for target detection.For high-definition remote sensing figure
For the application of picture, the detection of target is the important application of one.Ship Target in Marine remote sensing image has higher
Economic value and military value, it is a significant job that Ship Target, which be accurately positioned, based on remote sensing images.
The existing detection for Ship Target, the especially detection to Ship Target in high-definition picture, often by
To more noise jamming.Wherein, a kind of noise is the interference that complicated land atural object is introduced to surface vessel target detection.One
As by extracting the spatial relation characteristics of naval vessel hull and surrounding atural object, so that Ship Target be separated with land atural object to reach
The purpose of target detection.But it is due to that Ship Target and atural object relation are typically complex, for example, in actual conditions, naval vessel can
It can pull in shore to berth, easily obscure with atural object.Directly consider the spatial relationship of naval vessel and atural object, the effect of separating treatment is often not to the utmost
Such as people's will, so as to have impact on follow-up object detection results.
In addition, another noise is cloud layer introducing in natural scene.For this partial noise, its higher brightness value meeting
More false-alarm is brought to naval vessel testing result.Existing detection method often removes such using complex method and made an uproar
Sound, causes the detection to Ship Target in large format satellite image to need the larger computing resource of consumption and time.This method with
Based on two-dimensional fast fourier transform, by analyzing its spectrum signature, noise can be filtered out with fast speed and smaller cost,
The detection of Ship Target is completed, to obtain testing result and subsequent applications can be met.
The content of the invention
The invention aims to solve prior art because Ship Target and atural object are easily obscured and can not be effective
The problem of removing the larger computing resource of detection consumption and the time of Ship Target in noise and large format satellite image, and carry
A kind of Ship Target Detection method based on coastline data and spectrum analysis gone out.
Above-mentioned goal of the invention is achieved through the following technical solutions：
Step 1: obtaining the coastline vector data and high resolution ratio satellite remote-sensing image in same place, coastline is completed
The unification of the class data coordinate system of resolution ratio two of vector data engineer's scale and high resolution ratio satellite remote-sensing image, it is micro- in manual correction
After small details, land cultural noise in high resolution ratio satellite remote-sensing image is removed using land mask；Form background and be all ocean
View data be image after mask；
Step 2: carrying out piecemeal processing to the image after mask, image block is formed, image block is carried out in quick Fu of two dimension
Leaf transformation obtains the image block after Fourier transformation, judges whether the image block after Fourier transformation contains cloud layer noise；If Fu
In image block after leaf transformation contain cloud layer noise, then after adding the wave filter of Hamming window to Fourier transformation according to stencil design
Image block be filtered, obtain remove cloud layer noise after image；
Step 3: the binaryzation that gray level threshold segmentation is carried out to removing the image after cloud layer noise obtains binary image,
Seed point or seed region are detected in binary image；Obtained according to detecting seed point or seed region and carrying out region growing
Obtain naval vessel hull areas；
Wherein, seed point or seed region are detected carry out region growing and obtain the detailed process of naval vessel hull areas and be：
(1) obtain after seed region, the growing threshold using relative binary-state threshold less than 20% carries out eight neighborhood
Region growing；
(2) centered on the point in seed region, judge whether the value of the point in eight connected region is more than selected growing threshold；
If the value of the point in eight connected region is more than selected growing threshold, the value of the point in eight connected region is set to 1；
(3) repeat step (1) and (2) are until the result of the adjacent region growing twice in the region of eight neighborhood is consistent or repeats to walk
Suddenly untill (1) and (2) has carried out 20 times；Obtain the complete naval vessel hull areas that the region growing of eight neighborhood is obtained；
Step 4: being screened by the area and aspect ratio features of naval vessel hull areas, final high-resolution is obtained
Ship Target Detection result in satellite image.
Invention effect
The Ship Target Detection in high-resolution remote sensing image in the presence of a kind of different noises of solution of present invention proposition is asked
The method of topic.On the premise of coastline vector data and high-resolution remote sensing image is obtained, two class data are unified in same
Under coordinate system, enter line mask to land atural object using coastline data and reach to remove the purpose of a part of noise.And then more
Cloud layer noise is removed in simple ground substance environment, to reduce the detection false-alarm of noise like introducing, can finally pass through inspection
Survey method obtains the Ship Target in satellite image.
The present invention carries out mask process by using coastline data to land, although Ship Target and atural object easily generation are mixed
Confusing also can effectively remove noise.
Present invention seek to address that the Ship Target Detection problem in high-resolution remote sensing image in the presence of different noises.
On the premise of obtaining coastline vector data and high-resolution remote sensing image, two class data are unified under the same coordinate system, profit
Enter line mask to land atural object with coastline data and reach to remove the purpose of a part of noise.And then in relatively simple atural object ring
Cloud layer noise is removed in border, to reduce the detection false-alarm of noise like introducing, finally warship can be obtained by detection method
Ship target.
Land noise remove based on coastline data；
The premise for carrying out land noise remove is that the coastline vector data and high-resolution for having obtained same place are defended
Star remote sensing images, the unification of two class data coordinate systems are completed according to the resolution ratio of vector data engineer's scale and satellite image, in hand
After dynamic amendment minor detail, complete land mask to reach the purpose of land cultural noise removal.
Cloud layer noise remove based on spectrum signature；
Two-dimensional fast fourier transform is carried out to image block, for the image block with cloud layer noise, the frequency of its range value
Spectrum signature is there is obvious projection along one or two reference axis.Filtered according to the wave filter that stencil design adds Hamming window
Ripple, completes the removal of cloud layer noise.
Ship Target Detection based on Threshold segmentation and region growing；
After the removal for completing noise, row threshold division is entered to pending image, seed point or seed region is obtained.Then
The region growing for carrying out eight neighborhood obtains naval vessel hull areas, completes to screen finally by features such as the length-width ratios on naval vessel, obtains
Shown in final Ship Target Detection result, such as Figure 10 (a) and Figure 10 (b).
Brief description of the drawings
Fig. 1 is the Ship Target Detection method stream based on coastline data and spectrum analysis that embodiment one is proposed
Cheng Tu；
Fig. 2 (a) is the visualization general view schematic diagram for the coastline data that embodiment two is proposed；
Fig. 2 (b) is the local detail schematic diagram for the coastline data that embodiment two is proposed；
Fig. 3 (a) is that the result after the overlapping of the coastline vector data that embodiment two is proposed and satellite image is shown
It is intended to；
Fig. 3 (b) is the result schematic diagram after the line mask of being gone forward side by side to coastline vector refinement that embodiment two is proposed；
Fig. 4 (a) is the image block exemplary plot for the only cloud layer noise that embodiment four is proposed；
Fig. 4 (b) is feature schematic diagram of the image block for the only cloud layer noise that embodiment four is proposed on frequency domain；
Fig. 5 (a) is the cloud layer noise and the simultaneous image block exemplary plot of Ship Target that embodiment four is proposed；
Fig. 5 (b) is the cloud layer noise and the simultaneous image block of Ship Target of the proposition of embodiment four in frequency domain
On feature schematic diagram；
Fig. 6 (a) is the image block exemplary plot for the only Ship Target that embodiment four is proposed；
Fig. 6 (b) is feature schematic diagram of the image block for the only Ship Target that embodiment four is proposed on frequency domain；
(white portion is represented to be passed through the wave filter design template schematic diagram that Fig. 7 (a) proposes for embodiment five
The part of wave filter, the representative of black can not be by the part of wave filter)；
Fig. 7 (b) is the frequency response schematic diagram for the denoising filter that embodiment six is proposed；
Fig. 8 (a) is the example images figure removed before cloud layer noise that embodiment six is proposed；
Fig. 8 (b) is the example images figure removed after cloud layer noise that embodiment one is proposed；
Fig. 9 (a) is the exemplary plot of Ship Target image block 1 that embodiment one is proposed；
Fig. 9 (b) is the seed region schematic diagram after the binaryzation of Ship Target image block 1 that embodiment one is proposed；
Fig. 9 (c) is the exemplary plot of Ship Target image block 2 that embodiment one is proposed；
Fig. 9 (d) is the seed region schematic diagram after the binaryzation of Ship Target image block 2 that embodiment one is proposed
Figure 10 (a) is the final detection area schematic for the Ship Target image block 1 that embodiment eight is proposed；
Figure 10 (b) is the final detection area schematic for the Ship Target image block 2 that embodiment eight is proposed.
Embodiment
Embodiment one：A kind of Ship Target Detection based on coastline data and spectrum analysis of present embodiment
Method, is specifically what is prepared according to following steps：
Step 1: the land noise remove based on coastline data；Obtain the coastline vector data and height in same place
Definition satellite remote sensing images, are completed according to the resolution ratio of coastline vector data engineer's scale and high resolution ratio satellite remote-sensing image
The unification (completing to determine coastline vector data and high resolution ratio satellite remote-sensing image coordinate origin) of two class data coordinate systems,
After manual correction minor detail, land cultural noise in high resolution ratio satellite remote-sensing image is removed using land mask；Formed
Image do not have land clutter, background be all the image after the view data of ocean i.e. mask；
Step 2: the cloud layer noise remove based on spectrum signature；Piecemeal processing is carried out to the image after mask, image is formed
Block, carries out two-dimensional fast fourier transform to image block and obtains the image after the image block after Fourier transformation, Fourier transformation
The spectrum signature of the range value of block is that map of magnitudes has obvious projection along one or two reference axis of frequency domain, such as Fig. 4 (b) institutes
Show；Judge whether the image block after Fourier transformation contains cloud layer noise；Made an uproar if the image block after Fourier transformation contains cloud layer
Sound, the then wave filter for adding Hamming window according to stencil design is filtered to the image block after Fourier transformation, obtains removing cloud
Image after layer noise；The image of device filtering is shown in that Fig. 8 (b) completes the removal of cloud layer noise after filtering；Wherein, two-dimentional Fourier
What conversion was obtained is the data on frequency domain, and the image after it is visualized is shown in Figure of description 4 (b), now after Fourier transformation
Image block carry cloud layer noise；
Step 3: the Ship Target Detection based on Threshold segmentation and region growing；After the removal for completing cloud layer noise, to going
Except the binaryzation that the image after cloud layer noise carries out gray level threshold segmentation obtains binary image, detected in binary image
Seed point or seed region and the larger naval vessel region part of Sea background gray difference are that seed point or seed region are shown in Fig. 9
(a), (b), (c) and (d), according to detecting seed point or seed region carries out region growing and obtains naval vessel hull areas；
Wherein, seed point or seed region are detected carry out region growing and obtain the detailed process of naval vessel hull areas and be：
(1) obtain after seed region, the growing threshold using relative binary-state threshold less than 20% carries out eight neighborhood
Region growing；
(2) centered on the point in seed region, judge whether the value of the point in eight connected region (eight neighborhood) is more than and select
Growing threshold；If the value of the point in eight connected region (eight neighborhood) is more than selected growing threshold, by eight connected region (eight neighborhood)
The value of point is set to 1；If the value of the point in eight connected region (eight neighborhood) is less than selected growing threshold, any processing is not made；
(3) repeat step (1) and (2) (region growing for persistently carrying out eight neighborhood) until the region of eight neighborhood it is adjacent twice
Untill the result of region growing is consistent or repeat step (1) and (2) have carried out 20；The region growing of eight neighborhood obtains complete warship
Ship hull areas；
Step 4: being screened by the area and aspect ratio features of naval vessel hull areas, final high-resolution is obtained
Ship Target Detection result overall procedure is shown in Fig. 1 in satellite image.
Present embodiment effect：
The Ship Target inspection in high-resolution remote sensing image in the presence of a kind of different noises of solution of present embodiment proposition
The method of survey problem.On the premise of coastline vector data and high-resolution remote sensing image is obtained, two class data are unified in
Under the same coordinate system, enter line mask to land atural object using coastline data and reach to remove the purpose of a part of noise.And then
Cloud layer noise is removed in relatively simple ground substance environment, to reduce the detection false-alarm of noise like introducing, can finally be led to
The Ship Target crossed in detection method acquisition satellite image.
Present embodiment carries out mask process by using coastline data to land, although Ship Target is easily sent out with atural object
Life, which is obscured, also can effectively remove noise.
Present embodiment aims to solve the problem that the Ship Target Detection in the high-resolution remote sensing image in the presence of different noises is asked
Topic.On the premise of coastline vector data and high-resolution remote sensing image is obtained, two class data are unified in the same coordinate system
Under, enter line mask to land atural object using coastline data and reach to remove the purpose of a part of noise.And then relatively simple
Cloud layer noise is removed in ground substance environment, to reduce the detection false-alarm of noise like introducing, can finally pass through detection method
Obtain Ship Target.
Land noise remove based on coastline data；
The premise for carrying out land noise remove is that the coastline vector data and high-resolution for having obtained same place are defended
Star remote sensing images, the unification of two class data coordinate systems are completed according to the resolution ratio of vector data engineer's scale and satellite image, in hand
After dynamic amendment minor detail, complete land mask to reach the purpose of land cultural noise removal.
Cloud layer noise remove based on spectrum signature；
Two-dimensional fast fourier transform is carried out to image block, for the image block with cloud layer noise, the frequency of its range value
Spectrum signature is there is obvious projection along one or two reference axis.Filtered according to the wave filter that stencil design adds Hamming window
Ripple, completes the removal of cloud layer noise.
Ship Target Detection based on Threshold segmentation and region growing；
After the removal for completing noise, row threshold division is entered to pending image, seed point or seed region is obtained.Then
The region growing for carrying out eight neighborhood obtains naval vessel hull areas, completes to screen finally by features such as the length-width ratios on naval vessel, obtains
Shown in final Ship Target Detection result, such as Figure 10 (a) and Figure 10 (b).
Embodiment two：Present embodiment from unlike embodiment one：Obtained samely in step one
The coastline vector data and high resolution ratio satellite remote-sensing image of point, complete coastline vector data engineer's scale and high-resolution is defended
The unification of the class data coordinate system of resolution ratio two of star remote sensing images, after manual correction minor detail, is removed using land mask
Land cultural noise in high resolution ratio satellite remote-sensing image；Form background and be all the tool of the image after the view data of ocean i.e. mask
Body process is：
(1), data prepare；Base when the metadata record image that one panel height resolution remote sense satellite image is carried is obtained
This information；, it is necessary to which image provides the positional information shot for the masking method of land, typically represented with longitude and latitude；
Coastline data (see Fig. 2 (a) and Fig. 2 (b)) are provided in vector data mode；
(2), the intermediate-resolution seashore calculated using the disclosed U.S.National Oceanic and Atmospheric Administration of coastline data
The engineer's scale of line number evidence completes the unification of data coordinate system；
Determine that coastline vector data and high-resolution remote sensing image data belong to same geographical position according to longitude and latitude, and
And require that the directions of coastline vector data and high-resolution remote sensing image data is in and direct north aligned condition；
The important attribute of coastline vector data is engineer's scale, is completed when vector is superimposed with view data, it is necessary to ensure
Vector data has appropriate engineer's scale；Engineer's scale is converted with image resolution ratio by below equation：
Wherein PPI is per inch pixel count, and the PPI of general computer display is 72；0.0254 is English unit's (English
It is very little) and metric unit (rice) conversion constant；Vector data and view data quite well can be ensured according to above-mentioned formula,
Result in comparatively ideal data investigation effect；Fig. 3 (a) is shown in signal after superposition
(3), manual fine-tuning；Because the error of vector view data is inevitable, coastline vector data and height are completed
After the superposition of resolution remote sensing images data, the coastline that coastline vector data is indicated often is difficult to agree with image completely
Seashore, it is therefore desirable to be manually finely adjusted；
By coastline vector data and high-resolution remote sensing image data display in same data visualization window interface,
Vector data is made up of point, therefore the point element of coastline vector data is moved freely, by the way of artificial
Inaccurate corresponding coastline boundary point is adjusted on the coastline shown into high-resolution remote sensing image；
(4), land mask；After coastline in remote sensing images is indicated by coastline vector data, remove high
Land part in resolution remote sensing images, you can the image after output land mask, as a result sees Fig. 3 (b)；Wherein, after mask
Image eliminate great land atural object influenceed on Ship Target Detection, forms view data of the background for ocean.Other steps
Rapid and parameter is identical with embodiment one.
Embodiment three：Present embodiment from unlike embodiment one or two：Engineer's scale limit range
It is 1:10000 to 1:600000.Other steps and parameter are identical with embodiment one or two.
Embodiment four：Unlike one of present embodiment and embodiment one to three：It is right in step 2
Image after mask carries out piecemeal processing, forms image block, and carrying out two-dimensional fast fourier transform to image block obtains Fourier
Image block after conversion, judges whether the image block after Fourier transformation contains cloud layer noise；If the image after Fourier transformation
Block contains cloud layer noise, then the wave filter for adding Hamming window according to stencil design is filtered to the image block after Fourier transformation
Ripple, obtains removing the detailed process of the image after cloud layer noise：
(1), the spectrum sigtral response of cloud layer noise
Cloud layer noise is one of common noise in Marine remote sensing image, sees Fig. 4 (a), Fig. 5 (a)；The general color of cloud layer
Shallow, scope is big, edge is relatively obscured, and contrast of its larger gray value often with Sea background is more strong, is easily judged as
Doubtful naval vessel region, so as to increase the false-alarm probability of detection；
Piecemeal is carried out to the image after mask to be processed into the image block and length of the pixel of 512 pixels × 512 or wide be less than 512
Image block, is shown in Fig. 4 (a), 5 (a), 6 (a)；Two-dimensional fast fourier transform is carried out to image block；Obtain the figure after Fourier transformation
As block；
(2), in domain space, the amplitude of the image block frequency spectrum after Fourier transformation is calculated, and by after Fourier transformation
The amplitude of image block frequency spectrum be converted to log unit, be expressed as the map of magnitudes of domain space with decibel (dB) form；In frequency domain
Space draws map of magnitudes, sees Fig. 4 (b), 5 (b), 6 (b), in map of magnitudes it is apparent that cloud as noise in the presence of,
It is larger along one or two reference axis of frequency domain and its peripheral region amplitude that feature on frequency domain amplitude spectrum shows as map of magnitudes, display
Go out a kind of raised feature；Judge that the image block cloud layer noise after Fourier transformation whether there is, if it does, being made an uproar according to cloud layer
The feature of sound, designs filter template and adds Hamming window and filter out cloud layer noise from the image block after Fourier transformation.Its
Its step and parameter are identical with one of embodiment one to three.
Embodiment five：Unlike one of present embodiment and embodiment one to four：Judge that cloud layer is made an uproar
The detailed process that sound whether there is is：
(1), according to this feature of cloud layer noise, the image block after the Fourier transformation in domain space is calculated with frequency domain
X-axis and y-axis are respectively axis, extend the belt-like zone of 5 unit coordinate length formation respectively along axis both sides in space
Interior amplitude average value see Fig. 7 (a) differentiate as signal to noise ratio in component of signal；By four be divided into by frequency domain coordinates axle as
Amplitude average value in limit is used as noise component(s)；In domain space, the amplitude of frequency spectrum is converted into after logarithm Fourier transformation
The calculating of signal to noise ratio of image be expressed as：
Signal to noise ratio (dB)=component of signal-noise component(s)
For one snr threshold (15dB~25dB) of signal-to-noise ratio settings, judge whether the image block after Fourier transformation has
Cloud layer noise is present；Exist if the image block signal to noise ratio after Fourier transformation is more than the image block after threshold value, Fourier transformation
Cloud layer noise；
(2), according to the feature of cloud layer noise, design filter template and add Hamming window by cloud layer noise from Fourier become
Filtered out in image block after changing.Other steps and parameter are identical with one of embodiment one to four.
Embodiment six：Unlike one of present embodiment and embodiment one to five：Made an uproar according to cloud layer
The feature of sound, designs filter template and adds the tool that Hamming window filters out cloud layer noise from the image block after Fourier transformation
Body process is：
Determine after feature of the cloud layer noise on frequency domain amplitude spectrum, cloud layer noise is removed to rear using the method for frequency domain filtering
The influence of continuous naval vessel detection；
1) when being filtered to cloud layer noise, design filters out the filter template of cloud layer noise；
2) ringing effect directly will be introduced using ideal filter, therefore added on the Filtering Template that step (1) is obtained
Hamming window coefficient；Although plus cut-off frequency scope has expanded after hamming window, the influence compared to ringing effect to image, this
It is acceptable to plant error, wherein, coefficient ω (n) calculation formula of hamming window are：
Due to having divided the image into as 512 × 512 sizes, therefore length of window L=512, wherein, L=N+1；ω(n)
It is filter length, n wave filter stepping point sequences for filter coefficient, N；
3) after Filtering Template obtains the coefficient of hamming window, two dimension is designed by the coefficient and filter template of hamming window
There is limit for length's impact response filter i.e. FIR filter, Fig. 7 (b) is shown in the frequency response of wave filter；Obtain after FIR filter, in sky
Between convolution carried out in domain filter out cloud layer noise image；
4) due in practical application numerical computations necessarily introduce error, therefore directly do not filter out cloud layer noise in frequency domain；
The method of use is the pixel value that high resolution ratio satellite remote-sensing image is made after wave filter filters out cloud layer noise and the cloud layer filtered out
The pixel value of noise image subtracts each other, so as to obtain denoising image；Fig. 8 (a) is shown in result signal before and after denoising.Other steps and ginseng
Number is identical with one of embodiment one to five.
Embodiment seven：Unlike one of present embodiment and embodiment one to six：It is grey in step 3
Degree Threshold segmentation binarization method be specially：
More obvious region in naval vessel region is extracted using the method for gray level threshold segmentation；The two of intensity slicing will be carried out
Value threshold value is set as the 75%-85% of image maximum gradation value；The pixel value of high resolution ratio satellite remote-sensing image is more than or equal to
Binary-state threshold, is set to 1 in binary image；The pixel value of high resolution ratio satellite remote-sensing image is less than binary-state threshold,
0 is set in binary image.Other steps and parameter are identical with one of embodiment one to six.
Embodiment eight：Unlike one of present embodiment and embodiment one to seven：Lead in step 5
Cross the detailed process that the area and aspect ratio features of naval vessel hull areas are screened：
(1), with reference to image resolution ratio, hull area occupied on image in naval vessel has certain limitation, if therefore warship
The area of ship hull areas target bianry image be not located at reasonable interval (areal extent of reasonable interval be 200 to 800 pixels,
Not in this range regard as non-Ship Target), then by naval vessel hull areas object judgement be non-Ship Target, by non-warship
Ship target is rejected；Wherein, the areal extent of reasonable interval is 200 to 800 pixels；
(2), because naval vessel belongs to culture, the exclusive streamlined features in naval vessel to represent naval vessel mesh in bianry image
The length-width ratio for marking the minimum enclosed rectangle in region compares that non-Ship Target is more special, and its length-width ratio should also be in certain interval；
Judge in bianry image non-naval vessel hull areas area close to naval vessel hull areas target but resemblance by aspect ratio features
It is not belonging to the region on naval vessel；If the length-width ratio of doubtful Ship Target is in the range of 3 to 8, judge doubtful Ship Target as naval vessel
Hull areas target；If the length-width ratio of doubtful Ship Target is outside 3 to 8 scopes, judge doubtful Ship Target as non-naval vessel mesh
Mark；Final testing result is shown in Figure 10 (a) and Figure 10 (b).One of other steps and parameter and embodiment one to seven phase
Together.
Claims (8)
1. a kind of Ship Target Detection method based on coastline data and spectrum analysis, it is characterised in that one kind is based on coastline
Data and the Ship Target Detection method of spectrum analysis are specifically what is followed the steps below：
Step 1: obtaining the coastline vector data and high resolution ratio satellite remote-sensing image in same place, seashore line vector is completed
The unification of the class data coordinate system of resolution ratio two of ratio data chi and high resolution ratio satellite remote-sensing image is small thin in manual correction
After section, land cultural noise in high resolution ratio satellite remote-sensing image is removed using land mask；Form the figure that background is all ocean
It is the image after mask as data；
Step 2: carrying out piecemeal processing to the image after mask, image block is formed, two-dimentional fast Fourier is carried out to image block and is become
Get the image block after Fourier transformation in return, judge whether the image block after Fourier transformation contains cloud layer noise；If Fourier
Image block after conversion contains cloud layer noise, then adds the wave filter of Hamming window to the figure after Fourier transformation according to stencil design
As block is filtered, obtain removing the image after cloud layer noise；
Step 3: the binaryzation that gray level threshold segmentation is carried out to removing the image after cloud layer noise obtains binary image, two
Seed point or seed region are detected in value image；Warship is obtained according to detecting seed point or seed region and carrying out region growing
Ship hull areas；
Wherein, seed point or seed region are detected carry out region growing and obtain the detailed process of naval vessel hull areas and be：
(1) obtain after seed region, the growing threshold using relative binary-state threshold less than 20% carries out the region of eight neighborhood
Growth；
(2) centered on the point in seed region, judge whether the value of the point in eight connected region is more than selected growing threshold；If eight
The value of point in connected domain is more than selected growing threshold, and the value of the point in eight connected region is set into 1；
(3) repeat step (1) with (2) until the result of the adjacent region growing twice in the region of eight neighborhood is consistent or repeat step
(1) and untill (2) have carried out 20 times；Obtain the complete naval vessel hull areas that the region growing of eight neighborhood is obtained；
Step 4: being screened by the area and aspect ratio features of naval vessel hull areas, final high-resolution satellite is obtained
Ship Target Detection result in image.
2. a kind of Ship Target Detection method based on coastline data and spectrum analysis, its feature according to claim 1
It is：The coastline vector data and high resolution ratio satellite remote-sensing image in same place are obtained in step one, coastline arrow is completed
The unification of the class data coordinate system of resolution ratio two of ratio data chi and high resolution ratio satellite remote-sensing image is measured, it is small in manual correction
After details, land cultural noise in high resolution ratio satellite remote-sensing image is removed using land mask；Form background and be all ocean
View data is that the image detailed process after mask is：
(1), data prepare；Coastline data are provided in vector data mode；
(2), the intermediate-resolution seashore line number calculated using the disclosed U.S.National Oceanic and Atmospheric Administration of coastline data
According to engineer's scale complete data coordinate system unification；
Determine that coastline vector data and high-resolution remote sensing image data belong to same geographical position according to longitude and latitude, and will
The direction for asking coastline vector data and high-resolution remote sensing image data is in and direct north aligned condition；
Wherein, engineer's scale is converted with image resolution ratio by below equation：
In formula, PPI is per inch pixel count；0.0254 is the conversion constant of English unit and metric unit；
(3), manual fine-tuning；By coastline vector data and high-resolution remote sensing image data display in same data visualization window
In mouth interface, it will adjust into high-resolution remote sensing image aobvious without accurate corresponding coastline boundary point by the way of artificial
On the coastline shown；
(4), land mask；After coastline in remote sensing images is indicated by coastline vector data, high-resolution is removed
Land part in rate remote sensing images, you can the image after output land mask；Wherein, the image after mask is eliminated to naval vessel
Target detection influences great land atural object, forms the view data that background is ocean.
3. a kind of Ship Target Detection method based on coastline data and spectrum analysis, its feature according to claim 2
It is：Engineer's scale limit range is 1:10000 to 1:600000.
4. a kind of Ship Target Detection method based on coastline data and spectrum analysis, its feature according to claim 3
It is：Piecemeal processing is carried out to the image after mask in step 2, image block is formed, two-dimentional fast Fourier is carried out to image block
Conversion obtains the image block after Fourier transformation, judges whether the image block after Fourier transformation contains cloud layer noise；If in Fu
Image block after leaf transformation contains cloud layer noise, then according to stencil design add Hamming window wave filter to Fourier transformation after
Image block is filtered, and obtains removing the detailed process of the image after cloud layer noise：
(1) piecemeal, is carried out to the image after mask to be processed into the image block and length of the pixel of 512 pixels × 512 or wide be less than 512
Image block；Two-dimensional fast fourier transform is carried out to image block；Obtain the image block after Fourier transformation；
(2), in domain space, the amplitude of the image block frequency spectrum after Fourier transformation is calculated, and by the figure after Fourier transformation
As the amplitude of block frequency spectrum is converted to log unit, it is expressed as the map of magnitudes of domain space with decibel form；Drawn in domain space
Map of magnitudes, judges that the image block cloud layer noise after Fourier transformation whether there is, if it does, according to the feature of cloud layer noise,
Design filter template and add Hamming window and filter out cloud layer noise from the image block after Fourier transformation.
5. a kind of Ship Target Detection method based on coastline data and spectrum analysis, its feature according to claim 4
It is：Judge that the detailed process that cloud layer noise whether there is is：
(1), according to this feature of cloud layer noise, the image block after the Fourier transformation in domain space is calculated respectively with frequency domain
X-axis and y-axis are axis, extended respectively along axis both sides in the belt-like zone of 5 unit coordinate length formation in space
Amplitude average value differentiate as signal to noise ratio in component of signal；Amplitude in be divided into by frequency domain coordinates axle four quadrants is put down
Average is used as noise component(s)；In domain space, the amplitude of frequency spectrum is converted into the noise of the image after logarithm Fourier transformation
The calculating of ratio is expressed as：
Signal to noise ratio (dB)=component of signal-noise component(s)
For one snr threshold of signal-to-noise ratio settings, judge the image block after Fourier transformation whether with the presence of cloud layer noise；If
Image block signal to noise ratio after Fourier transformation is more than threshold value, then the image block after Fourier transformation has cloud layer noise；
(2), according to the feature of cloud layer noise, design filter template and add Hamming window by cloud layer noise after Fourier transformation
Image block in filter out.
6. a kind of Ship Target Detection method based on coastline data and spectrum analysis, its feature according to claim 5
It is：According to the feature of cloud layer noise, design filter template and add Hamming window by cloud layer noise after Fourier transformation
The detailed process filtered out in image block is：
1) when being filtered to cloud layer noise, design filters out the filter template of cloud layer noise；
2) in step 1) on obtained Filtering Template plus hamming window coefficient；Wherein, coefficient ω (n) calculation formula of hamming window are：
<mrow>
<mi>&omega;</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mn>0.54</mn>
<mo>-</mo>
<mn>0.46</mn>
<mi>c</mi>
<mi>o</mi>
<mi>s</mi>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mi>&pi;</mi>
<mfrac>
<mi>n</mi>
<mi>N</mi>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
Due to having divided the image into as 512 × 512 sizes, therefore length of window L=512, wherein, L=N+1；ω (n) is filter
Ripple device coefficient, N are filter length, n wave filter stepping point sequences；
3) it is FIR filter to design the long impact response filter of two-dimensional finite by the coefficient and filter template of hamming window；Obtain
Obtain after FIR filter, convolution is carried out in the spatial domain and filters out cloud layer noise image；
4) because after wave filter filters out cloud layer noise, the pixel value for making high resolution ratio satellite remote-sensing image is made an uproar with the cloud layer filtered out
The pixel value of acoustic image subtracts each other, so as to obtain denoising image.
7. a kind of Ship Target Detection method based on coastline data and spectrum analysis, its feature according to claim 6
It is：The binarization method of gray level threshold segmentation is specially in step 3：
The binary-state threshold for carrying out intensity slicing is set as to the 75%-85% of image maximum gradation value；High-resolution satellite is distant
The pixel value for feeling image is more than or equal to binary-state threshold, and 1 is set in binary image；The picture of high resolution ratio satellite remote-sensing image
Element value is less than binary-state threshold, and 0 is set in binary image.
8. a kind of Ship Target Detection method based on coastline data and spectrum analysis, its feature according to claim 7
It is：The detailed process screened in step 4 by the area and aspect ratio features of naval vessel hull areas：
(1) if, the area of naval vessel hull areas target bianry image be not located at reasonable interval, naval vessel hull areas target is sentenced
Break as non-Ship Target, non-Ship Target is rejected；Wherein, the areal extent of reasonable interval is 200 to 800 pixels；
(2) if, the length-width ratio of doubtful Ship Target in the range of 3 to 8, judge doubtful Ship Target as naval vessel hull areas mesh
Mark；If the length-width ratio of doubtful Ship Target is outside 3 to 8 scopes, judge doubtful Ship Target as non-Ship Target.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510594428.8A CN105303526B (en) | 2015-09-17 | 2015-09-17 | A kind of Ship Target Detection method based on coastline data and spectrum analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510594428.8A CN105303526B (en) | 2015-09-17 | 2015-09-17 | A kind of Ship Target Detection method based on coastline data and spectrum analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105303526A CN105303526A (en) | 2016-02-03 |
| CN105303526B true CN105303526B (en) | 2017-11-07 |
Family
ID=55200750
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510594428.8A Active CN105303526B (en) | 2015-09-17 | 2015-09-17 | A kind of Ship Target Detection method based on coastline data and spectrum analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105303526B (en) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106405491B (en) * | 2016-08-29 | 2018-11-30 | 成都川美新技术股份有限公司 | Unmanned plane based on software radio monitors system |
| CN106408624B (en) * | 2016-09-30 | 2019-07-23 | 浙江大学 | A kind of fast automatic stacking method of coastline data suitable for more projecting and remote sensing images |
| CN106447738B (en) * | 2016-09-30 | 2019-08-30 | 浙江大学 | It is a kind of that using GDAL and SHAPELIB, rapid superposition shows seashore line method on remote sensing images |
| CN106709426B (en) * | 2016-11-29 | 2020-04-14 | 上海航天测控通信研究所 | Ship target detection method based on infrared remote sensing image |
| CN107346545A (en) * | 2017-05-22 | 2017-11-14 | 沈阳工业大学 | Improved confinement growing method for the segmentation of optic cup image |
| CN107240084B (en) * | 2017-06-14 | 2021-04-02 | 湘潭大学 | Method and device for removing rain from single image |
| CN109961065B (en) * | 2017-12-25 | 2021-05-25 | 中国科学院沈阳自动化研究所 | Sea surface ship target detection method |
| CN108508427B (en) * | 2018-07-04 | 2020-07-07 | 烟台雷奥电子科技有限公司 | Sea ice area detection method, device and equipment based on navigation radar |
| CN109064430B (en) * | 2018-08-07 | 2020-10-09 | 中国人民解放军陆军炮兵防空兵学院 | Cloud removing method and system for aerial region cloud-containing image |
| CN109859178B (en) * | 2019-01-18 | 2020-11-03 | 北京航空航天大学 | FPGA-based infrared remote sensing image real-time target detection method |
| CN110649957B (en) * | 2019-08-26 | 2023-08-01 | 西安空间无线电技术研究所 | Method, system and medium for reducing transmission error rate of satellite-borne microwave radiometer slip ring |
| CN111833329A (en) * | 2020-07-14 | 2020-10-27 | 中国电子科技集团公司第五十四研究所 | Manual evidence judgment auxiliary method for large remote sensing image |
| CN111860336B (en) * | 2020-07-21 | 2022-02-11 | 西北工业大学 | High-resolution remote sensing image inclined ship target detection method based on position sensing |
| CN112085656B (en) * | 2020-08-31 | 2023-12-05 | 厦门蓝海天信息技术有限公司 | Ocean mode data shoreline extension method, system, mobile terminal and storage medium |
| CN115082359B (en) * | 2022-07-27 | 2022-11-22 | 中国科学院空天信息创新研究院 | Synchronous orbit optical satellite geometric fine correction method based on coastline data |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101604383A (en) * | 2009-07-24 | 2009-12-16 | 哈尔滨工业大学 | A kind of method for detecting targets at sea based on infrared image |
| CN102663348A (en) * | 2012-03-21 | 2012-09-12 | 中国人民解放军国防科学技术大学 | Marine ship detection method in optical remote sensing image |
| CN103810487A (en) * | 2014-01-24 | 2014-05-21 | 深圳大学 | Method and system for target detection and identification of aerial ocean images |
| CN103886285A (en) * | 2014-03-11 | 2014-06-25 | 武汉大学 | Method for detecting optical remote-sensing image ship with assistance of prior geographic information |
| CN104240257A (en) * | 2014-09-30 | 2014-12-24 | 中国人民解放军国防科学技术大学 | SAR (synthetic aperture radar) image naval ship target identification method based on change detection technology |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8311301B2 (en) * | 2010-12-10 | 2012-11-13 | Carestream Health, Inc. | Segmenting an organ in a medical digital image |
-
2015
- 2015-09-17 CN CN201510594428.8A patent/CN105303526B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101604383A (en) * | 2009-07-24 | 2009-12-16 | 哈尔滨工业大学 | A kind of method for detecting targets at sea based on infrared image |
| CN102663348A (en) * | 2012-03-21 | 2012-09-12 | 中国人民解放军国防科学技术大学 | Marine ship detection method in optical remote sensing image |
| CN103810487A (en) * | 2014-01-24 | 2014-05-21 | 深圳大学 | Method and system for target detection and identification of aerial ocean images |
| CN103886285A (en) * | 2014-03-11 | 2014-06-25 | 武汉大学 | Method for detecting optical remote-sensing image ship with assistance of prior geographic information |
| CN104240257A (en) * | 2014-09-30 | 2014-12-24 | 中国人民解放军国防科学技术大学 | SAR (synthetic aperture radar) image naval ship target identification method based on change detection technology |
Non-Patent Citations (1)
| Title |
|---|
| SAR 图像港口目标检测研究;刘成皓;《中国优秀硕士学位论文全文数据库 信息科技辑》;20130215(第2期);正文第3-44页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105303526A (en) | 2016-02-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105303526B (en) | A kind of Ship Target Detection method based on coastline data and spectrum analysis | |
| CN101727662B (en) | SAR image nonlocal mean value speckle filtering method | |
| CN102930512B (en) | Based on the underwater picture Enhancement Method of HSV color space in conjunction with Retinex | |
| CN109636771B (en) | Flight target detection method and system based on image processing | |
| Bhatti et al. | Advanced color edge detection using clifford algebra in satellite images | |
| CN105046258B (en) | A kind of object detection method and device of small target detection sonar image | |
| CN111027497B (en) | Weak and small target rapid detection method based on high-resolution optical remote sensing image | |
| CN109427055A (en) | The remote sensing images surface vessel detection method of view-based access control model attention mechanism and comentropy | |
| CN106709914A (en) | SAR image ship detection false alarm eliminating method based on two-stage DEM sea-land reservoir | |
| CN104715474B (en) | High resolution synthetic aperture radar linearity building object detecting method based on Based On Method of Labeling Watershed Algorithm | |
| CN108564602A (en) | Airplane detection method based on airport remote sensing image | |
| CN110852992B (en) | Ship trail detection method based on top hat transformation and Radon transformation | |
| US20180144478A1 (en) | Method and system for sea background modeling and suppression on high-resolution remote sensing sea images | |
| CN104021553B (en) | A kind of sonar image object detection method based on pixel layering | |
| CN110245600B (en) | Unmanned aerial vehicle road detection method for self-adaptive initial quick stroke width | |
| CN107194355A (en) | A kind of utilization orientation derivative constructs the method for detecting infrared puniness target of entropy contrast | |
| CN115272876A (en) | Remote sensing image ship target detection method based on deep learning | |
| CN112150482B (en) | Sea-land segmentation method based on elevation information and normalized sea water index | |
| CN117079117B (en) | Underwater image processing and target identification method and device, storage medium and electronic equipment | |
| CN103455986B (en) | Random noise point detecting method based on fractional order differential gradient | |
| CN112489108A (en) | Method and device for inverting and reconstructing underwater surface area of open-sea coral reef | |
| CN105551029A (en) | Multi-spectral remote sensing image-based ship detection method | |
| CN110046607A (en) | A kind of unmanned aerial vehicle remote sensing image board house or building materials test method based on deep learning | |
| RU2308055C2 (en) | Method for forming of image of sea vessel contour according to radar surveillances | |
| CN109918968A (en) | A kind of Ship Target Detection method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |