WO2018000252A1 - 高分辨遥感海洋图像的海背景建模与抑制的方法及系统 - Google Patents

高分辨遥感海洋图像的海背景建模与抑制的方法及系统 Download PDF

Info

Publication number
WO2018000252A1
WO2018000252A1 PCT/CN2016/087716 CN2016087716W WO2018000252A1 WO 2018000252 A1 WO2018000252 A1 WO 2018000252A1 CN 2016087716 W CN2016087716 W CN 2016087716W WO 2018000252 A1 WO2018000252 A1 WO 2018000252A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
sub
image block
analyzed
ocean
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.)
Ceased
Application number
PCT/CN2016/087716
Other languages
English (en)
French (fr)
Inventor
裴继红
王荔霞
谢维信
杨烜
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen University
Original Assignee
Shenzhen University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen University filed Critical Shenzhen University
Priority to PCT/CN2016/087716 priority Critical patent/WO2018000252A1/zh
Priority to US15/857,551 priority patent/US10497128B2/en
Publication of WO2018000252A1 publication Critical patent/WO2018000252A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/20—Image enhancement or restoration using local operators
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/143—Segmentation; Edge detection involving probabilistic approaches, e.g. Markov random field [MRF] modelling
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/168—Segmentation; Edge detection involving transform domain methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
    • G06V10/431—Frequency domain transformation; Autocorrelation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20021—Dividing image into blocks, subimages or windows
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20048—Transform domain processing
    • G06T2207/20056—Discrete and fast Fourier transform, [DFT, FFT]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30236—Traffic on road, railway or crossing

Definitions

  • the present invention relates to the field of image processing technologies, and in particular, to a method and system for modeling and suppressing sea background of high resolution remote sensing ocean image.
  • the target information is included, but also the seawater background information of the ocean surface around the target.
  • the seawater background is affected by various natural factors such as sea surface wind, wind direction, surge, and environmental humidity. There are many forms of expression, such as wind waves, surges, waves, eddies, foams, and so on.
  • modeling seawater background and suppressing seawater background is a better approach.
  • This method uses structural primitives to find spatially repetitive regions with similar characteristics, namely seawater background regions, and removes these regions from the image to extract the target region.
  • the results depend on the choice of structural primitives. How to choose better structural primitives is the key and difficult point of this method.
  • This method first selects a probabilistic model (such as Gaussian model, K-distribution model, etc.) that best describes the statistical distribution characteristics of the sea surface background spatial image of the ocean remote sensing image, and secondly according to the sea surface.
  • the spatial gray scale of the background estimates the parameters of the distribution model.
  • the model probability judgment is performed on the gray value of each pixel in the ocean remote sensing image containing the target, thereby segmenting Ship target area.
  • the method is suitable for the sea surface background, and the sea surface clutter can be selected by fitting a suitable distribution model. However, for images with more complex background clutter, the distribution model is often not well defined. The accuracy of the segmentation of the target area.
  • This method first uses the fractal theory and technology to perform multi-scale decomposition of the fractional dimension of the image, and then divides the two according to the difference between the fractal dimension of the sea background region and the target region, so as to detect and Extract the target area.
  • the single scale or constant fractal dimension is difficult to distinguish between the sea surface and the target area.
  • Method based on visual saliency model The method first generates a visual saliency map by feature extraction, saliency calculation and saliency map fusion, and secondly finds a more significant visual object in the generated saliency map and extracts the region. To achieve detection of the target area. This method introduces a variety of features, which can better segment the target area and the sea surface background. However, due to the many features, their reasonable choices are not well evaluated.
  • the existing background model based on image space gray distribution can not describe the background clutter in the image well when the sea background is complex, and can not well carry out the sea surface background of satellite remote sensing ocean image.
  • the fitting method makes the existing target detection method of satellite remote sensing ocean image based on spatial gray distribution model have the problems of high false alarm rate and low detection accuracy.
  • the object of the present invention is to provide a method and system for sea background modeling and suppression of high resolution remote sensing ocean image, aiming at solving the goal of satellite remote sensing ocean image based on spatial gray distribution model in the prior art.
  • the detection method has the problems of high false alarm rate and low detection accuracy.
  • the invention provides a sea background modeling and suppression method for high resolution remote sensing ocean image, characterized in that the method comprises:
  • Cutting step pre-processing and block cutting the remote sensing image, and rough-chowing each of the cut sub-image blocks, and selecting a sub-image block set capable of analyzing the ocean image;
  • Classification step further classifying all sub-image blocks in the sub-image block set of the analyzable ocean image into a pure seawater image block subset and a non-pure seawater image block subset;
  • Calculating step performing a two-dimensional discrete Fourier transform on each sub-image block in the set of sub-image blocks of the analyzable ocean image, obtaining a corresponding spectrogram, and calculating a corresponding amplitude spectrum from the spectrogram;
  • a modeling step determining a sub-image block to be analyzed in a set of sub-image blocks of the analyzable ocean image, centering on the sub-image block to be analyzed, and finding the sub-image block to be analyzed in the sub-set of pure seawater image blocks All the pure seawater image blocks around to construct a sea background amplitude spectrum Gaussian probability model map;
  • Suppression step constructing a sea background suppression filter by using the amplitude spectrum Gaussian probability model map, and performing sea background suppression on the sub-image block to be analyzed.
  • the method further includes:
  • Cycling step determining a next sub-image block to be analyzed, and performing sea background suppression on the new sub-image block to be analyzed by repeating the modeling step and the suppressing step until sea background suppression of all sub-image blocks is completed.
  • the step of suppressing specifically includes:
  • the Gaussian background suppression filter is used to perform frequency domain filtering on the sub-image block to be analyzed, and the filtered spectral image is subjected to two-dimensional inverse Fourier transform to obtain a sub-image block subjected to background suppression.
  • the present invention also provides a system for sea background modeling and suppression of high resolution remote sensing ocean images, the system comprising:
  • a cutting module configured to perform pre-processing and block cutting on the remote sensing image, and perform rough classification on each of the cut sub-image blocks, and select a sub-image block set that can analyze the ocean image;
  • a classification module configured to further classify all sub-image blocks in the sub-image block set of the analyzable ocean image into a pure seawater image block subset and a non-pure seawater image block subset;
  • a calculation module for each sub-image block in the set of sub-image blocks of the analyzable ocean image Performing a two-dimensional discrete Fourier transform to obtain a corresponding spectrogram, and calculating a corresponding amplitude spectrum from the spectrogram;
  • a modeling module configured to determine a sub-image block to be analyzed in a set of sub-image blocks that can analyze the ocean image, and to find the to-be-analyzed object in the sub-set of pure seawater image blocks centering on the sub-image block to be analyzed All pure seawater image blocks around the image block to construct a sea background amplitude spectrum Gaussian probability model map;
  • a suppression module configured to construct a sea background suppression filter by using the amplitude spectrum Gaussian probability model map, and perform sea background suppression on the sub-image block to be analyzed.
  • the system further comprises:
  • a loop module configured to determine a next sub-image block to be analyzed, and perform sea background suppression on the new sub-image block to be analyzed by repeating the modeling step and the suppressing step until the sea background of all sub-image blocks is completed inhibition.
  • the suppression module is specifically configured to:
  • the Gaussian background suppression filter is used to perform frequency domain filtering on the sub-image block to be analyzed, and the filtered spectral image is subjected to two-dimensional inverse Fourier transform to obtain a sub-image block subjected to background suppression.
  • the technical solution provided by the present invention better solves the problem that the sea surface background model of the satellite remote sensing ocean image used in the prior art cannot accurately describe and suppress the sea surface background clutter in the image, resulting in unstable ocean remote sensing image target detection result.
  • the problem of the invention can greatly improve the detection accuracy rate and reduce the false alarm rate by first suppressing the ocean background and then detecting the target.
  • FIG. 1 is a schematic diagram of sea background modeling and suppression of high resolution remote sensing ocean image according to an embodiment of the present invention.
  • FIG. 2a is a schematic diagram of an example of a sub-image block to be analyzed according to an embodiment of the present invention
  • Figure 2b is an amplitude spectrum corresponding to Figure 2a in an embodiment of the present invention.
  • FIG. 3 is a detailed detailed flowchart of the suppression step shown in FIG. 1 according to an embodiment of the present invention
  • 4a is a schematic diagram of an ideal sea background suppression filter according to an embodiment of the present invention.
  • Figure 4b is an enlarged view of a partial area of Figure 4a according to an embodiment of the present invention.
  • Figure 4c is a three-dimensional view of Figure 4b in accordance with an embodiment of the present invention.
  • FIG. 5a is a schematic diagram of a Gaussian sea background suppression filter according to an embodiment of the present invention.
  • Figure 5b is an enlarged view of a partial area of Figure 5a according to an embodiment of the present invention.
  • Figure 5c is a three-dimensional view of Figure 5b in accordance with one embodiment of the present invention.
  • FIG. 6 is a schematic diagram showing the result of filtering the image block to be analyzed shown in FIG. 2a by using the Gaussian sea background suppression filter shown in FIG. 5a according to an embodiment of the present invention
  • FIG. 7 is a schematic diagram showing the internal structure of a system 10 for modeling and suppressing sea background of high resolution remote sensing ocean image according to an embodiment of the present invention.
  • the main idea of the sea background modeling and suppression method for the optical satellite high-resolution remote sensing ocean image proposed by the present invention is to establish a frequency domain statistical model of the local sea background for the remote sensing ocean image. Based on this, a background suppression filter is built to suppress the sea background.
  • the specific implementation process includes: a cutting step: preprocessing and block cutting the remote sensing image, and roughly classifying each sub-image block that is cut out, and selecting a sub-image block set capable of analyzing the ocean image; classification step: pairable analysis All sub-image blocks in the sub-image block set of the marine image are further classified into a pure sea water image block sub-set and a non-pure sea water image block sub-set; a calculation step: a sub-image of the analyzable ocean image Each sub-image block in the block set performs a two-dimensional discrete Fourier transform to obtain a corresponding spectrogram, and the corresponding amplitude spectrum is calculated from the spectrogram; modeling step: determining in the sub-image block set of the analyzable ocean image A sub-image block to be analyzed is centered on the sub-image block to be analyzed, and a sea background amplitude spectrum Gaussian probability model is constructed by finding all the pure seawater image blocks around the sub-image block to be analyzed in the
  • the sea background modeling and suppression method of the high-resolution remote sensing ocean image provided by the invention can greatly improve the detection accuracy rate and reduce the false alarm rate by first suppressing the ocean background and then detecting the target. .
  • FIG. 1 is a flowchart of a method for modeling and suppressing sea background of high resolution remote sensing ocean image according to an embodiment of the present invention.
  • step S1 the cutting step: pre-processing and block-cutting the remote sensing image, and performing rough classification on each of the cut sub-image blocks, and selecting a sub-image block set capable of analyzing the ocean image.
  • the remote sensing image is subjected to preprocessing such as sea-land segmentation and thick cloud region detection; the pre-processed remote sensing image is subjected to block cutting; and the cut sub-image blocks are roughly classified into the analyzable ocean image sub-image.
  • preprocessing such as sea-land segmentation and thick cloud region detection
  • block cutting such as block cutting
  • the cut sub-image blocks are roughly classified into the analyzable ocean image sub-image.
  • the set Sa and the non-analyzable ocean image subgraph set Sn are examples of preprocessing such as sea-land segmentation and thick cloud region detection
  • the sea-land segmentation of the remote sensing image refers to segmenting the terrestrial and island regions in the remote sensing image from the ocean image, so that the detection of the marine target only needs to be performed in the marine region.
  • the global coastline database can be used in conjunction with automatic learning to deal with the land and sea segmentation problem.
  • some marine regions are often covered by thick clouds, in the thick cloud region of these remote sensing images. It is impossible to detect marine targets.
  • the thick cloud region in the remote sensing image can be effectively detected by some existing thick cloud detection algorithms.
  • the sea-land segmentation method and the thick cloud detection method will not be described again.
  • it is assumed that terrestrial and sea in the remote sensing image to be processed have been segmented from the ocean image by sea-land segmentation and thick cloud detection methods. Island area, as well as thick cloud areas.
  • the format of the high-resolution remote sensing image is generally large, the coverage is large, the number of pixels is large, and the amount of data is large.
  • the ocean background has similarities in local regions, while the background similarity is reduced in two regions that are far apart.
  • the size of the sub-block is 256x256 or 512x512.
  • the sub-picture in a sub-image of the divided block, if the sum of the area occupied by the land area and the thick cloud area exceeds a certain ratio r1 of the total area of the sub-image, the sub-picture is called an unanalyzable ocean sub-picture. .
  • the set of images composed of all non-analyzable ocean subgraphs is called the non-analyzable ocean image subgraph set Sn.
  • the above-mentioned proportional coefficient r1 can generally be selected as 50%, or 75%, according to the accuracy requirement.
  • the sub-image set composed of the remaining sub-image blocks is called the analyzable ocean image sub-graph set Sa.
  • all subsequent methods are performed for images in the analyzable ocean image subgraph set Sa.
  • step S2 the classifying step further classifies all the sub-image blocks in the sub-image block set of the analyzable ocean image into a pure sea water image block sub-set and a non-pure sea water image block sub-set.
  • each sub-image in the analyzable ocean image sub-graph set Sa is further roughly classified, and the analyzable ocean image sub-graph set Sa is divided into two sub-sets: a pure sea water image block sub-set Sw and a non- Pure seawater image block sub-set So.
  • the sub-image block in the ocean image sub-graph set Sa may be analyzed as an image of pure sea water, or may be an image block containing a target such as a ship.
  • the established ocean background is a statistical model of the spectral domain
  • the accuracy of the background image modeling of the ocean image is affected; The smaller the area, the accuracy of the ocean image background modeling will not have a significant impact.
  • the area occupied by the target here is Small in the present invention means that the target occupies less than 1% of the total area of the sub-image block in the sub-image block.
  • some techniques can well complete the rough classification of pure seawater and non-pure seawater for sub-image blocks in the set Sa, such as the classification method based on histogram non-single peak detection, etc., which can analyze the ocean image from the above.
  • the set of images Sa is better distinguished from the sub-image blocks in which the larger targets are included.
  • These sub-image blocks containing larger targets constitute a sub-set So of non-pure sea water image blocks, and other image blocks in the set Sa except So constitute a pure sea water image block sub-set Sw.
  • the specific division method will not be described again.
  • the above-mentioned pure seawater image block subset Sw may contain some non-pure seawater sub-image blocks with a small target area.
  • the non-pure seawater image block generally includes: an image block containing a part of land or island, an image block containing a large target to be detected, and an image block including a part of a thick cloud area.
  • step S3 the calculating step is: performing a two-dimensional discrete Fourier transform on each sub-image block in the set of sub-image blocks of the analyzable ocean image to obtain a corresponding spectrogram, and calculating a corresponding amplitude spectrum from the spectrogram.
  • a two-dimensional Fourier transform is performed on each block sub-image f k in the block image set Sa, and a spectrogram F k corresponding to each sub-image block is obtained, and the spectrogram is calculated.
  • Amplitude spectrum A k is calculated.
  • each sub-image in the block image set Sa is NxN, wherein the kth sub-image is f k (x, y), and (x, y) is the pixel point space coordinate of the image, 0 ⁇ x ⁇ N-1, 0 ⁇ y ⁇ N-1.
  • the spectrum of the two-dimensional discrete space Fourier transform F k (u, v), 0 ⁇ u ⁇ N-1, 0 ⁇ v ⁇ N-1 is calculated by the formula (1):
  • R k (u, v) and I k (u, v) are graphs of the real part composition and the imaginary part of F k (u, v), respectively.
  • step S4 the modeling step is: determining a block of the sub-image block to be analyzed in the set of sub-image blocks of the analyzable ocean image, centering on the sub-image block to be analyzed, and finding the All pure seawater image blocks around the sub-image block to be analyzed are used to construct a sea background amplitude spectrum Gaussian probability model map.
  • a block of image f i to be analyzed is determined in the image set Sa. Focusing on the image block f i , all pure seawater image blocks near f i are found in the pure sea water image subset Sw to form a subset Sw i , and the amplitude spectrum Gaussian probability of the sea background is constructed by using the image blocks in Sw i Model diagram G i .
  • a block of image to be analyzed in the block image set Sa is f i (x, y) whose size is NxN
  • the spectrum of the two-dimensional discrete space Fourier transform is F i (u, v)
  • the amplitude spectrum is A i (u, v), 0 ⁇ u ⁇ N-1, 0 ⁇ v ⁇ N-1.
  • FIG. 2a is an example of a sub-image block to be analyzed according to an embodiment of the present invention
  • FIG. 2b is a corresponding amplitude spectrum diagram thereof.
  • the vicinity of the above f i refers to a rectangular or circular neighborhood centered on the image block f i , and the side length N of the cut image block is in units, and the neighborhood radius can be selected from 2N to 5N. In the embodiment of the present invention, a circular neighborhood is selected, and the neighborhood radius is selected as 3N.
  • the Sw i sub-set contains J-block pure seawater image blocks, where the j-th pure seawater sub-image block is f j (x, y), 0 ⁇ x ⁇ N-1, 0 ⁇ y ⁇ N-1, and second
  • the spectrum of the dimensionally discrete space Fourier transform is F j (u, v)
  • the amplitude spectrum is A j (u, v), 0 ⁇ u ⁇ N-1, 0 ⁇ v ⁇ N-1.
  • m i (u, v) and [ ⁇ i (u, v)] 2 are the mean values of the image block amplitude spectra in the pure seawater image block subset Sw i near the image block f i (x, y) to be analyzed, respectively.
  • the graph and the variance map, called m i (u, v) and [ ⁇ i (u, v)] 2 are the sea background amplitude spectrum Gaussian probability model map G i of the image block f i (x, y) to be analyzed.
  • step S5 the suppressing step is to construct a sea background suppression filter using the amplitude spectrum Gaussian probability model map, and perform sea background suppression on the sub-image block to be analyzed.
  • the sea background suppression filter B i is constructed using the amplitude spectrum A i of the image block f i and the amplitude spectrum probability model G i of the sea background, and the image is subjected to background suppression using B i .
  • step S5 further includes the following steps:
  • step S51 the sea background amplitude spectrum Gaussian probability model map constructed by the modeling step is used to calculate a filtered Markov distance map of the sub-image block to be analyzed.
  • the filtered Markov distance map Q(u, v) of the image block f i to be analyzed is calculated using the background probability model G i of the image block f i calculated in step S4.
  • step S52 the filtered Markov distance map is used to obtain the rationality of the sub-image block to be analyzed.
  • the sea background suppression filter Think of the sea background suppression filter.
  • the filtering step S51 Markov calculated to be analyzed from the image block f i of FIG. Q (u, v), to give the design image to be analyzed over the sea background suppression filter block f i ⁇ i (u , v).
  • the ideal sea background suppression filter ⁇ i (u, v) of the image block f i can be calculated by the formula (7).
  • T in the formula (7) is a constant, which is called a decision threshold.
  • step S53 the Gaussian Sea Background Suppression Filter is further designed using the ideal sea background suppression filter.
  • the ideal sea background suppression filter ⁇ i (u, v) will produce a "ringing" effect in practical applications, thereby affecting the suppression effect of the ocean background.
  • the present invention smoothes the ideal sea background suppression filter obtained in step S52 by using a Gaussian kernel function, and obtains a Gaussian sea background suppression filter B i (u, v) which can overcome the "ringing" effect.
  • the specific calculation formula is shown in formula (8):
  • the Gaussian smooth function template H d (u,v) can be calculated by equation (9):
  • ⁇ 1;
  • W is a normalized constant associated with a smoothing function, calculated by equation (10):
  • step S54 the image block to be analyzed is subjected to frequency domain filtering by using the Gaussian sea background suppression filter, and the filtered spectrum image is subjected to two-dimensional inverse Fourier transform to obtain a sub-image block subjected to background suppression.
  • sea background suppression filter B i to be analyzed for frequency domain filtering of the image block f i, and the frequency spectrum of the filtered image by the two-dimensional inverse Fourier transform, to give the sub-image blocks after background suppression f ti .
  • the Fourier spectrum of the sub-image block f i (x, y) to be analyzed is F i (u, v)
  • the Gaussian background suppression filter is B i (u, v)
  • the background suppression filter calculation formula is as shown in formula (11):
  • FIG. 4, FIG. 5 and FIG. 6 show an example provided by an embodiment of the present invention.
  • 4a is an ideal seawater background suppression filter for the calculated image block sea area of FIG. 2a
  • FIG. 4b is an enlarged view of a partial area of the white square in FIG. 4a
  • FIG. 4c is a three-dimensional view of FIG. 4b.
  • Fig. 5a is a Gaussian seawater background suppression filter in the sea of Fig. 2a of the calculated image block
  • Fig. 5b is an enlarged view of a partial area where the white square in Fig. 5a is located
  • Fig. 5c is a three-dimensional view of Fig. 5b.
  • Figure 6 is a graph of the results of background suppression filtering of Figure 2a using the filter of Figure 5a and conversion to the spatial domain.
  • the method for modeling and suppressing sea background of a high-resolution satellite remote sensing ocean image further includes:
  • Cycling step determining a next block of sub-images to be analyzed, and performing sea background suppression on the new sub-image block to be analyzed by repeating the modeling step and the suppressing step until all sub-image blocks are completed Sea background suppression.
  • the invention provides a sea background modeling and suppression method for high resolution remote sensing ocean image, which better solves the sea surface background model of satellite remote sensing ocean image used in the prior art because it cannot describe and suppress the image well
  • the embodiment of the present invention further provides a system 10 for modeling and suppressing sea background of high resolution remote sensing ocean image, which mainly comprises:
  • the cutting module 11 is configured to perform pre-processing and block cutting on the remote sensing image, and perform rough classification on each of the cut sub-image blocks, and select a sub-image block set that can analyze the ocean image;
  • a classification module 12 configured to further classify all sub-image blocks in the sub-image block set of the analyzable ocean image into a pure seawater image block subset and a non-pure seawater image block subset;
  • the calculation module 13 is configured to perform a two-dimensional discrete Fourier transform on each sub-image block in the set of sub-image blocks of the analyzable ocean image, to obtain a corresponding spectrogram, and calculate a corresponding amplitude spectrum from the spectrogram;
  • the modeling module 14 is configured to determine a sub-image block to be analyzed in the sub-image block set of the analyzable ocean image, and focus on the sub-image block to be analyzed, and find the to-be-analyzed in the pure seawater image block subset All pure seawater image blocks around the sub-image block to construct a sea background amplitude spectrum Gaussian probability model map;
  • the suppression module 15 is configured to construct a sea background suppression filter by using the amplitude spectrum Gaussian probability model map, and perform sea background suppression on the sub-image block to be analyzed.
  • the system 10 for modeling and suppressing sea background of high-resolution remote sensing ocean image provided by the invention can greatly improve the detection accuracy and reduce the virtualness by first suppressing the ocean background and then detecting the target. police rate.
  • FIG. 7 a schematic structural diagram of a system 10 for modeling and suppressing sea background of high resolution remote sensing ocean image according to an embodiment of the present invention is shown.
  • a system for sea background modeling and suppression of high resolution remote sensing ocean images 10
  • the cutting module 11, the classification module 12, the calculation module 13, the modeling module 14, and the suppression module 15 are to be included.
  • the cutting module 11 is configured to perform pre-processing and block cutting on the remote sensing image, and perform rough classification on each of the cut sub-image blocks, and select a sub-image block set that can analyze the ocean image.
  • the specific cutting method of the cutting module 11 is referred to the related description in the foregoing step S1, and the repeated description is not repeated here.
  • the classification module 12 is further configured to classify all sub-image blocks in the sub-image block set of the analyzable ocean image into a pure seawater image block subset and a non-pure seawater image block subset.
  • the calculation module 13 is configured to perform a two-dimensional discrete Fourier transform on each sub-image block in the set of sub-image blocks of the analyzable ocean image to obtain a corresponding spectrogram, and calculate a corresponding amplitude spectrum from the spectrogram.
  • the modeling module 14 is configured to determine a sub-image block to be analyzed in the sub-image block set of the analyzable ocean image, and focus on the sub-image block to be analyzed, and find the to-be-analyzed in the pure seawater image block subset All pure seawater image blocks around the sub-image block are used to construct a sea background amplitude spectrum Gaussian probability model map.
  • the specific modeling method of the modeling module 14 is referred to the related description in the foregoing step S4, and the repeated description is not repeated here.
  • the suppression module 15 is configured to construct a sea background suppression filter by using the amplitude spectrum Gaussian probability model map, and perform sea background suppression on the sub-image block to be analyzed.
  • the suppression module 15 is specifically configured to:
  • the Gaussian background suppression filter is used to perform frequency domain filtering on the sub-image block to be analyzed, and the filtered spectral image is subjected to two-dimensional inverse Fourier transform to obtain a sub-image block subjected to background suppression.
  • system 10 for sea background modeling and suppression of high resolution remote sensing ocean image provided by the present invention further includes:
  • a loop module configured to determine a next sub-image block to be analyzed, and perform sea background suppression on the new sub-image block to be analyzed by repeating the modeling step and the suppressing step until the sea background of all sub-image blocks is completed inhibition.
  • the system 10 for sea background modeling and suppression of high-resolution remote sensing ocean image provided by the invention better solves the sea surface model model of the optical satellite remote sensing ocean image used in the prior art because it cannot describe and suppress well
  • the sea surface background clutter in the image leads to the problem that the target detection result of the ocean remote sensing image is unstable.
  • the invention can greatly improve the detection accuracy rate and reduce the virtual state by first suppressing the ocean background and then detecting the target. police rate.
  • each unit included is only divided according to functional logic, but is not limited to the above division, as long as the corresponding function can be implemented; in addition, the specific name of each functional unit is also They are only used to facilitate mutual differentiation and are not intended to limit the scope of the present invention.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Remote Sensing (AREA)
  • Astronomy & Astrophysics (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Radar Systems Or Details Thereof (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

本发明提供一种高分辨遥感海洋图像的海背景建模与抑制的方法，其中，所述方法包括：切割步骤、分类步骤、计算步骤、建模步骤以及抑制步骤。本发明还提供一种高分辨遥感海洋图像的海背景建模与抑制的系统。本发明提供的技术方案通过先对海洋背景进行抑制然后再对目标进行检测的这种方式，可以较大地提高检测正确率以及降低虚警率。

Description

高分辨遥感海洋图像的海背景建模与抑制的方法及系统 技术领域
本发明涉及图像处理技术领域，尤其涉及一种高分辨遥感海洋图像的海背景建模与抑制的方法及系统。
背景技术
在光学卫星遥感海洋图像中，不但包含有目标信息，而且还包含了目标周围海洋表面的海水背景信息，海水背景受海面风力、风向、浪涌、环境湿度等多种自然因素的影响，在图像中的表现形态多种多样，如风浪、浪涌、浪花、漩涡、泡沫等等。为了有效地检测出遥感海洋图像中的目标，对海水背景建模并进行海水背景抑制是一种较好的途径。
目前，现有技术中提出的针对光学卫星遥感海洋图像的海水背景建模和抑制的方法主要有以下几种：
一、基于数学形态学的方法：该方法使用结构基元寻找特性相似的空间重复性区域，即海水背景区域，并将这些区域从图像中去除，从而提取出目标区域。利用数学形态学的方法进行检测时，其结果依赖于结构基元的选择，如何选择较好的结构基元既是该方法的关键，也是难点。
二、基于图像空间灰度统计分布模型的方法：该方法首先选择一种最能描述海洋遥感图像海面背景空间灰度统计分布特性的概率模型(如高斯模型、K分布模型等)，其次根据海面背景的空间灰度对该分布模型的参数进行估计，最后利用这个海水背景空间灰度统计分布模型，对含有目标的海洋遥感图像中的每个像素点灰度值进行模型概率判断，从而分割出舰船目标区域。该方法对于海面背景比较平静的情况，海面杂波可以选择较为合适的分布模型来拟合。但是，对于背景杂波较为复杂的图像，其分布模型往往无法很好地确定，从而影 响目标区域分割的准确性。
三、基于分形模型的方法：该方法首先利用分形理论和技术对图像进行分数维度的多尺度分解，其次根据海面背景区域和目标区域在分形维数的差异对这两者进行分割，从而检测和提取出目标区域。但是实际图像中受背景复杂度、随机噪声、成像质量等的影响，单一的尺度或恒定的分形维数很难区分出海面背景和目标区域。
四、基于视觉显著度模型的方法：该方法首先通过特征提取、显著度计算和显著图融合来生成视觉显著图，其次在生成的显著图中寻找较为显著的视觉对象，并将该区域提取出来，实现目标区域的检测。该方法引入了多种特征，能够将目标区域和海面背景较好地分割开来，但由于特征较多，它们的合理选择却并没很好的评定方法。
综上所述，目前已有的基于图像空间灰度分布的背景模型，当海面背景较为复杂时无法很好地描述图像中的背景杂波，无法很好地对卫星遥感海洋图像的海面背景进行拟合，从而使得现有的基于空间灰度分布模型的卫星遥感海洋图像的目标检测方法存在虚警率高、检测正确率低的问题。
发明内容
有鉴于此，本发明的目的在于提供一种高分辨遥感海洋图像的海背景建模与抑制的方法及系统，旨在解决现有技术中针对基于空间灰度分布模型的卫星遥感海洋图像的目标检测方法存在虚警率高、检测正确率低的问题。
本发明提出一种高分辨遥感海洋图像的海背景建模与抑制的方法，其特征在于，所述方法包括：
切割步骤：对遥感图像进行预处理和分块切割，并将切割出的各个子图像块进行粗分类，选出可分析海洋图像的子图像块集合；
分类步骤：对可分析海洋图像的子图像块集合中的所有子图像块进一步分类为纯海水图像块子集合以及非纯海水图像块子集合；
计算步骤：对可分析海洋图像的子图像块集合中的每一个子图像块进行二维离散傅立叶变换，得到对应的频谱图，并由频谱图计算出对应的幅度谱图；
建模步骤：在可分析海洋图像的子图像块集合中确定一块待分析子图像块，以该待分析子图像块为中心，在纯海水图像块子集合中通过找出该待分析子图像块周围的所有纯海水图像块来构建海背景幅度谱高斯概率模型图；
抑制步骤：利用所述幅度谱高斯概率模型图构建海背景抑制滤波器，并对该待分析子图像块进行海背景抑制。
优选的，所述方法还包括：
循环步骤：确定下一块待分析子图像块，并通过重复所述建模步骤和所述抑制步骤来对新的待分析子图像块进行海背景抑制，直至完成所有子图像块的海背景抑制。
优选的，所述抑制步骤具体包括：
利用所述建模步骤构建出的所述海背景幅度谱高斯概率模型图，计算该待分析子图像块的滤波马氏距离图；
利用所述滤波马氏距离图设计得到该待分析子图像块的理想海背景抑制滤波器；
利用所述理想海背景抑制滤波器进一步设计高斯海背景抑制滤波器；
利用所述高斯海背景抑制滤波器对该待分析子图像块进行频域滤波，并将滤波后的频谱图像通过二维傅里叶反变换得到经过背景抑制的子图像块。
另一方面，本发明还提供一种高分辨遥感海洋图像的海背景建模与抑制的系统，所述系统包括：
切割模块，用于对遥感图像进行预处理和分块切割，并将切割出的各个子图像块进行粗分类，选出可分析海洋图像的子图像块集合；
分类模块，用于对可分析海洋图像的子图像块集合中的所有子图像块进一步分类为纯海水图像块子集合以及非纯海水图像块子集合；
计算模块，用于对可分析海洋图像的子图像块集合中的每一个子图像块进 行二维离散傅立叶变换，得到对应的频谱图，并由频谱图计算出对应的幅度谱图；
建模模块，用于在可分析海洋图像的子图像块集合中确定一块待分析子图像块，以该待分析子图像块为中心，在纯海水图像块子集合中通过找出该待分析子图像块周围的所有纯海水图像块来构建海背景幅度谱高斯概率模型图；
抑制模块，用于利用所述幅度谱高斯概率模型图构建海背景抑制滤波器，并对该待分析子图像块进行海背景抑制。
优选的，所述系统还包括：
循环模块，用于确定下一块待分析子图像块，并通过重复所述建模步骤和所述抑制步骤来对新的待分析子图像块进行海背景抑制，直至完成所有子图像块的海背景抑制。
优选的，所述抑制模块具体用于：
利用所述建模模块构建出的所述海背景幅度谱高斯概率模型图，计算该待分析子图像块的滤波马氏距离图；
利用所述滤波马氏距离图设计得到该待分析子图像块的理想海背景抑制滤波器；
利用所述理想海背景抑制滤波器进一步设计高斯海背景抑制滤波器；
利用所述高斯海背景抑制滤波器对该待分析子图像块进行频域滤波，并将滤波后的频谱图像通过二维傅里叶反变换得到经过背景抑制的子图像块。
本发明提供的技术方案较好地解决了现有技术所采用的卫星遥感海洋图像的海面背景模型由于无法很好地描述和抑制图像中的海面背景杂波而导致海洋遥感图像目标检测结果不稳定的问题，本发明通过先对海洋背景进行抑制然后再对目标进行检测的这种方式，可以较大地提高检测正确率以及降低虚警率。
附图说明
图1为本发明一实施方式中高分辨遥感海洋图像的海背景建模与抑制的方 法流程图；
图2a为本发明一实施方式中提供的一个待分析子图像块示例图；
图2b为本发明一实施方式中与图2a对应的幅度谱图；
图3为本发明一实施方式中图1所示的抑制步骤的具体详细流程图；
图4a为本发明一实施方式中提供的一个理想海背景抑制滤波器示意图；
图4b为本发明一实施方式中图4a的局部区域放大图；
图4c为本发明一实施方式中图4b的三维图；
图5a为本发明一实施方式中提供的一个高斯海背景抑制滤波器示意图；
图5b为本发明一实施方式中图5a的局部区域放大图；
图5c为本发明一实施方式中图5b的三维图；
图6为本发明一实施方式中使用图5a所示的高斯海背景抑制滤波器对图2a所示的待分析子图像块进行滤波后的结果示意图；
图7为本发明一实施方式中高分辨遥感海洋图像的海背景建模与抑制的系统10的内部结构示意图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
为了解决现有技术存在的问题，本发明提出的光学卫星高分辨遥感海洋图像的海背景建模与抑制的方法的主要思想是：对遥感海洋图像建立局部的海背景的频域统计模型，在此基础上建立背景抑制滤波器对海背景进行抑制。具体实现过程包括：切割步骤：对遥感图像进行预处理和分块切割，并将切割出的各个子图像块进行粗分类，选出可分析海洋图像的子图像块集合；分类步骤：对可分析海洋图像的子图像块集合中的所有子图像块进一步分类为纯海水图像块子集合以及非纯海水图像块子集合；计算步骤：对可分析海洋图像的子图像 块集合中的每一个子图像块进行二维离散傅立叶变换，得到对应的频谱图，并由频谱图计算出对应的幅度谱图；建模步骤：在可分析海洋图像的子图像块集合中确定一块待分析子图像块，以该待分析子图像块为中心，在纯海水图像块子集合中通过找出该待分析子图像块周围的所有纯海水图像块来构建海背景幅度谱高斯概率模型图；抑制步骤：利用所述幅度谱高斯概率模型图构建海背景抑制滤波器，并对该待分析子图像块进行海背景抑制。
本发明提供的一种高分辨遥感海洋图像的海背景建模与抑制的方法通过先对海洋背景进行抑制然后再对目标进行检测的这种方式，可以较大地提高检测正确率以及降低虚警率。
以下将对本发明所提供的一种高分辨遥感海洋图像的海背景建模与抑制的方法进行详细说明。
请参阅图1，为本发明一实施方式中高分辨遥感海洋图像的海背景建模与抑制的方法流程图。
在步骤S1中，切割步骤：对遥感图像进行预处理和分块切割，并将切割出的各个子图像块进行粗分类，选出可分析海洋图像的子图像块集合。
在本实施方式中，对遥感图像进行海陆分割、厚云区域检测等预处理；对预处理后的遥感图像进行分块切割；将切割出的各个子图像块粗分类为可分析海洋图像子图集合Sa以及不可分析海洋图像子图集合Sn。
在本实施方式中，对遥感图像的海陆分割是指将遥感图像中的陆地和岛屿区域从海洋图像中分割出来，这样对海洋目标的检测就只需要在海洋区域中进行。在本实施方式中，可以使用全球海岸线数据库配合自动学习等方法来处理海陆分割问题，另外，在得到的海洋遥感图像中，往往有一些海洋区域被厚云覆盖，在这些遥感图像的厚云区域中是无法进行海洋目标检测的。通过目前已存在的一些厚云检测的算法可以将遥感图像中的厚云区域有效检测出来。关于海陆分割方法以及厚云检测方法将不再赘述。在本发明中，假设已通过海陆分割、以及厚云检测方法从海洋图像中分割出了待处理的遥感图像中的陆地和海 岛区域，以及厚云区域。
在本实施方式中，由于高分辨遥感图像的幅面一般都比较大，表现在覆盖的范围大，像素数多、数据量大。在海洋遥感图像中，海洋背景在局部区域具有相似性，而对相距比较远的两个区域，背景的相似性会变小。为了建立准确的海洋背景模型，有效地抑制背景、检测目标，需要对大幅面遥感图像进行分块处理。一般子块的大小选为256ⅹ256或512ⅹ512大小比较合适。在图像分块时，可以进行无交叠区域的均匀分块，也可以进行有交叠区域的均匀分块。为了更便于目标的检测，采用在水平相邻子块之间、以及垂直相邻子块之间均交叠50％的方式进行图像分块较为合适。
在本实施方式中，在已分块的某一子图像中，若陆地区域和厚云区域所占面积之和超过该子图像总面积一定比例r1时，该子图称为不可分析海洋子图。所有不可分析海洋子图组成的图像集合称为不可分析海洋图像子图集合Sn。在实际使用中，上述的比例系数r1一般可根据精度需要选为50％，或75％。对所有已分块的子图像，剔除在集合Sn中的所有子图像块后，剩余的子图像块构成的子图像集合称为可分析海洋图像子图集合Sa。在本发明中，后续的所有方法都是针对可分析海洋图像子图集合Sa中的图像进行的。
在步骤S2中，分类步骤：对可分析海洋图像的子图像块集合中的所有子图像块进一步分类为纯海水图像块子集合以及非纯海水图像块子集合。
在本实施方式中，对可分析海洋图像子图集合Sa中的每一块子图像进一步粗分类，将可分析海洋图像子图集合Sa划分为两个子集合：即纯海水图像块子集合Sw以及非纯海水图像块子集合So。
在本实施方式中，在步骤S2中，可分析海洋图像子图集合Sa中的子图像块可能是纯海水的图像，也可能是包含舰船等目标的图像块。在本发明中，由于所建立的海洋背景是一种频谱域的统计模型，若在子图像块中包含所占面积较大的目标，则会影响海洋图像背景建模的准确度；而若目标所占面积较小，则对海洋图像背景建模的准确度不会产生显著影响。此处所述目标所占面积较 小在本发明中是指目标在子图像块中所占面积小于子图像块总面积的1％。目前已有一些技术可以较好地完成对集合Sa中的子图像块进行纯海水与非纯海水的粗分类，如基于直方图非单峰检测的分类方法等，可以从上述可分析海洋图像子图集合Sa中较好地区分出其中含有较大目标的子图像块。这些含有较大目标的子图像块构成了非纯海水图像块子集合So，而集合Sa中除So的其它图像块则构成了纯海水图像块子集合Sw。具体的划分方法不再赘述。
需要指出的是，上述所属纯海水图像块子集合Sw中，可能会含有一些目标面积较小的非纯海水子图像块。而其中的非纯海水图像块中则一般包括：含有部分陆地或岛屿的图像块、含有较大的待检测目标的的图像块以及包含有部分厚云区的图像块等。
在本发明的后面步骤中，假定已获得了可分析海洋图像子图集合Sa，及该集合的两个子集合：纯海水图像块子集合Sw、以及非纯海水图像块子集合So。
在步骤S3中，计算步骤：对可分析海洋图像的子图像块集合中的每一个子图像块进行二维离散傅立叶变换，得到对应的频谱图，并由频谱图计算出对应的幅度谱图。
在本实施方式中，对分块图像集合Sa中的每一块分块子图像fk进行二维傅里叶变换，得到每一块子图像块对应的频谱图Fk，以及由频谱图计算出的幅度谱图Ak。
假设在分块图像集合Sa中的每一块子图像的大小为NⅹN，其中的第k个子图像为fk(x,y)，(x,y)为图像的像素点空间坐标，0≤x≤N-1，0≤y≤N-1。则其二维离散空间傅立叶变换的频谱图Fk(u,v)，0≤u≤N-1，0≤v≤N-1由公式(1)计算得到：
Figure PCTCN2016087716-appb-000001
相应的幅度谱Ak(u,v)由公式(2)计算得到：
Figure PCTCN2016087716-appb-000002
其中，Rk(u,v)，Ik(u,v)分别为Fk(u,v)的实部组成的图和虚部组成的图。
在步骤S4中，建模步骤：在可分析海洋图像的子图像块集合中确定一块待分析子图像块，以该待分析子图像块为中心，在纯海水图像块子集合中通过找出该待分析子图像块周围的所有纯海水图像块来构建海背景幅度谱高斯概率模型图。
在本实施方式中，在图像集合Sa中确定一块待分析的图像块fi。以图像块fi为中心，在纯海水图像子集合Sw中找出在fi附近的所有纯海水图像块，构成子集合Swi，用Swi中的图像块构建海背景的幅度谱高斯概率模型图Gi。
假设在分块图像集合Sa中的一块待分析图像块为fi(x,y)，其大小为NⅹN，其二维离散空间傅立叶变换的频谱图为Fi(u,v)，幅度谱为Ai(u,v)，0≤u≤N-1，0≤v≤N-1。
图2a是本发明实施例给出的一个待分析子图像块示例，图2b是其对应的幅度谱图。
以图像块fi为中心，在纯海水图像子集Sw中找出在fi附近的所有纯海水图像块，构成子集合Swi。上述所述fi附近是指以图像块fi为中心的一个矩形或圆形邻域，以前述已切割的图像块的边长N为单位，邻域半径可选取2N至5N。在本发明实施例中选用圆形邻域，且邻域半径选为3N。假设Swi子集合中含有J块纯海水图像块，其中第j个纯海水子图像块为fj(x,y)，0≤x≤N-1，0≤y≤N-1，其二维离散空间傅立叶变换的频谱图为Fj(u,v)，幅度谱图为Aj(u,v)，0≤u≤N-1，0≤v≤N-1。
进一步，利用纯海水图像块子集合Swi中的所有幅度谱图Aj(u,v)，构建待分析图像fi(x,y)处海域的海背景的幅度谱高斯概率模型图Gi，该模型的概率密度函数可表示为：
Figure PCTCN2016087716-appb-000003
其中，
Figure PCTCN2016087716-appb-000004
Figure PCTCN2016087716-appb-000005
上述mi(u,v)和[σi(u,v)]2分别是待分析图像块fi(x,y)附近的纯海水图像块子集合Swi中的图像块幅度谱的均值图和方差图，称mi(u,v)和[σi(u,v)]2为待分析图像块fi(x,y)的海背景幅度谱高斯概率模型图Gi。通过上述公式(4)和公式(5)，在图像块的每一个频率点(u,v)处建立了一个海背景的高斯分布统计模型，其均值与方差分别为mi(u,v)和[σi(u,v)]2。
在步骤S5中，抑制步骤：利用所述幅度谱高斯概率模型图构建海背景抑制滤波器，并对该待分析子图像块进行海背景抑制。
在本实施方式中，使用图像块fi的幅度谱图Ai以及海背景的幅度谱概率模型Gi构建海背景抑制滤波器Bi，使用Bi对图像进行背景抑制。
具体地，如图3所示，步骤S5又包括以下步骤：
在步骤S51中，利用所述建模步骤构建出的所述海背景幅度谱高斯概率模型图，计算该待分析子图像块的滤波马氏距离图。
在本实施方式中，利用步骤S4计算出的图像块fi的背景概率模型Gi，计算待分析图像块fi的滤波马氏距离图Q(u,v)。
滤波马氏距离图Q(u,v)的计算公式为公式(6)：
Figure PCTCN2016087716-appb-000006
在步骤S52中，利用所述滤波马氏距离图设计得到该待分析子图像块的理 想海背景抑制滤波器。
在本实施方式中，利用步骤S51计算得到的待分析图像块fi的滤波马氏距离图Q(u,v)，设计得到待分析图像块fi的理想海背景抑制滤波器Ωi(u,v)。
在本实施方式中，图像块fi的理想海背景抑制滤波器Ωi(u,v)可由公式(7)计算得到，
Figure PCTCN2016087716-appb-000007
其中，在公式(7)中的T是一个常数，称为判决阈值。T的值一般可以取值为区间[1,9]之间的实数。在本发明实施例中，取值T＝9。
在步骤S53中，利用所述理想海背景抑制滤波器进一步设计高斯海背景抑制滤波器。
在本实施方式中，利用设计出的理想海背景抑制滤波器Ωi(u,v)，进一步设计高斯海背景抑制滤波器Bi(u,v)。
在本实施方式中，理想海背景抑制滤波器Ωi(u,v)在实际应用中会产生“振铃”效应，从而影响海洋背景的抑制效果。本发明利用高斯核函数对步骤S52中得到的理想海背景抑制滤波器进行平滑，得到可以克服“振铃”效应的高斯海背景抑制滤波器Bi(u,v)。具体计算公式如公式(8)所示：
Bi(u,v)＝Ωi(u,v)*Hd(u,v)             (8)其中，在公式(8)中，公式中的符号*表示数字图像处理中的二维卷积运算。Hd(u,v)表示以(u,v)为中心的大小为(2d+1)×(2d+1)的高斯型平滑函数模板，d为正整数。在本发明实施例中，取d＝1。
高斯型平滑函数模板Hd(u,v)可由公式(9)计算得到：
Figure PCTCN2016087716-appb-000008
其中，在公式(9)中，(u,v)取值为范围-d≤u≤d,-d≤v≤d中的整数；参数λ是控制平滑强度的常数，一般可以取λ＝d。在本发明实施例中，取λ＝1；W 是一个与平滑函数有关的归一化的常数，由公式(10)计算得到：
Figure PCTCN2016087716-appb-000009
在步骤S54中，利用所述高斯海背景抑制滤波器对该待分析子图像块进行频域滤波，并将滤波后的频谱图像通过二维傅里叶反变换得到经过背景抑制的子图像块。
在本实施方式中，利用海背景抑制滤波器Bi进行待分析图像块fi的频域滤波，并将滤波后的频谱图像通过二维傅里叶反变换，得到经过背景抑制的子图像块fti。
在本实施方式中，如前所示，待分析子图像块fi(x,y)的傅里叶频谱图为Fi(u,v)，高斯海背景抑制滤波器为Bi(u,v)，背景抑制滤波计算公式如公式(11)所示：
Fti(u,v)＝Fi(u,v)Bi(u,v)                   (11)
计算背景抑制后的子图像块fti(x,y)的二维傅里叶反变换如公式(12)所示：
Figure PCTCN2016087716-appb-000010
图4、图5、图6给出了本发明实施例提供的一个实例。其中，图4a是计算出的图像块图2a海域的理想海水背景抑制滤波器，图4b是图4a中的白色方框所在的局部区域放大图，图4c是图4b的三维图。图5a是计算出的图像块图2a海域的高斯海水背景抑制滤波器，图5b是图5a中的白色方框所在的局部区域放大图，图5c是图5b三维图。图6是使用图5a滤波器对图2a进行背景抑制滤波后转换到空间域的结果图。
除此之外，本发明提供的一种高分辨卫星遥感海洋图像的海背景建模与抑制的方法，进一步还包括：
循环步骤：确定下一块待分析子图像块，并通过重复所述建模步骤和所述抑制步骤来对新的待分析子图像块进行海背景抑制，直至完成所有子图像块的 海背景抑制。
本发明提供的一种高分辨遥感海洋图像的海背景建模与抑制的方法，较好地解决了现有技术所采用的卫星遥感海洋图像的海面背景模型由于无法很好地描述和抑制图像中的海面背景杂波而导致海洋遥感图像目标检测结果不稳定的问题，本发明通过先对海洋背景进行抑制然后再对目标进行检测的这种方式，可以较大地提高检测正确率以及降低虚警率。
本发明具体实施方式还提供一种高分辨遥感海洋图像的海背景建模与抑制的系统10，主要包括：
切割模块11，用于对遥感图像进行预处理和分块切割，并将切割出的各个子图像块进行粗分类，选出可分析海洋图像的子图像块集合；
分类模块12，用于对可分析海洋图像的子图像块集合中的所有子图像块进一步分类为纯海水图像块子集合以及非纯海水图像块子集合；
计算模块13，用于对可分析海洋图像的子图像块集合中的每一个子图像块进行二维离散傅立叶变换，得到对应的频谱图，并由频谱图计算出对应的幅度谱图；
建模模块14，用于在可分析海洋图像的子图像块集合中确定一块待分析子图像块，以该待分析子图像块为中心，在纯海水图像块子集合中通过找出该待分析子图像块周围的所有纯海水图像块来构建海背景幅度谱高斯概率模型图；
抑制模块15，用于利用所述幅度谱高斯概率模型图构建海背景抑制滤波器，并对该待分析子图像块进行海背景抑制。
本发明提供的一种高分辨遥感海洋图像的海背景建模与抑制的系统10，通过先对海洋背景进行抑制然后再对目标进行检测的这种方式，可以较大地提高检测正确率以及降低虚警率。
请参阅图7，所示为本发明一实施方式中高分辨遥感海洋图像的海背景建模与抑制的系统10的结构示意图。
在本实施方式中，高分辨遥感海洋图像的海背景建模与抑制的系统10，主 要包括切割模块11、分类模块12、计算模块13、建模模块14和抑制模块15。
切割模块11，用于对遥感图像进行预处理和分块切割，并将切割出的各个子图像块进行粗分类，选出可分析海洋图像的子图像块集合。
在本实施方式中，切割模块11的具体的切割方法请参阅前述步骤S1中的相关描述，在此就不做重复描述。
分类模块12，用于对可分析海洋图像的子图像块集合中的所有子图像块进一步分类为纯海水图像块子集合以及非纯海水图像块子集合。
在本实施方式中，分类模块12的具体分类方法请参阅前述步骤S2中的相关描述，在此就不做重复描述。
计算模块13，用于对可分析海洋图像的子图像块集合中的每一个子图像块进行二维离散傅立叶变换，得到对应的频谱图，并由频谱图计算出对应的幅度谱图。
在本实施方式中，计算模块13的具体计算方法请参阅前述步骤S3中的相关描述，在此就不做重复描述。
建模模块14，用于在可分析海洋图像的子图像块集合中确定一块待分析子图像块，以该待分析子图像块为中心，在纯海水图像块子集合中通过找出该待分析子图像块周围的所有纯海水图像块来构建海背景幅度谱高斯概率模型图。
在本实施方式中，建模模块14的具体建模方法请参阅前述步骤S4中的相关描述，在此就不做重复描述。
抑制模块15，用于利用所述幅度谱高斯概率模型图构建海背景抑制滤波器，并对该待分析子图像块进行海背景抑制。
在本实施方式中，抑制模块15具体用于：
利用所述建模模块构建出的所述海背景幅度谱高斯概率模型图，计算该待分析子图像块的滤波马氏距离图；
利用所述滤波马氏距离图设计得到该待分析子图像块的理想海背景抑制滤波器；
利用所述理想海背景抑制滤波器进一步设计高斯海背景抑制滤波器；
利用所述高斯海背景抑制滤波器对该待分析子图像块进行频域滤波，并将滤波后的频谱图像通过二维傅里叶反变换得到经过背景抑制的子图像块。
在本实施方式中，抑制模块15的具体抑制方法请参阅前述步骤S5中的相关描述，在此就不做重复描述。
除此之外，本发明提供的一种高分辨遥感海洋图像的海背景建模与抑制的系统10，进一步还包括：
循环模块，用于确定下一块待分析子图像块，并通过重复所述建模步骤和所述抑制步骤来对新的待分析子图像块进行海背景抑制，直至完成所有子图像块的海背景抑制。
本发明提供的一种高分辨遥感海洋图像的海背景建模与抑制的系统10，较好地解决了现有技术所采用的光学卫星遥感海洋图像的海面背景模型由于无法很好地描述和抑制图像中的海面背景杂波而导致海洋遥感图像目标检测结果不稳定的问题，本发明通过先对海洋背景进行抑制然后再对目标进行检测的这种方式，可以较大地提高检测正确率以及降低虚警率。
值得注意的是，上述实施例中，所包括的各个单元只是按照功能逻辑进行划分的，但并不局限于上述的划分，只要能够实现相应的功能即可；另外，各功能单元的具体名称也只是为了便于相互区分，并不用于限制本发明的保护范围。
另外，本领域普通技术人员可以理解实现上述各实施例方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成，相应的程序可以存储于一计算机可读取存储介质中，所述的存储介质，如ROM/RAM、磁盘或光盘等。
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等，均应包含在本发明的保护范围之内。

Claims (6)

  1. 一种高分辨遥感海洋图像的海背景建模与抑制的方法，其特征在于，所述方法包括：
    切割步骤：对遥感图像进行预处理和分块切割，并将切割出的各个子图像块进行粗分类，选出可分析海洋图像的子图像块集合；
    分类步骤：对可分析海洋图像的子图像块集合中的所有子图像块进一步分类为纯海水图像块子集合以及非纯海水图像块子集合；
    计算步骤：对可分析海洋图像的子图像块集合中的每一个子图像块进行二维离散傅立叶变换，得到对应的频谱图，并由频谱图计算出对应的幅度谱图；
    建模步骤：在可分析海洋图像的子图像块集合中确定一块待分析子图像块，以该待分析子图像块为中心，在纯海水图像块子集合中通过找出该待分析子图像块周围的所有纯海水图像块来构建海背景幅度谱高斯概率模型图；
    抑制步骤：利用所述幅度谱高斯概率模型图构建海背景抑制滤波器，并对该待分析子图像块进行海背景抑制。
  2. 如权利要求1所述的高分辨遥感海洋图像的海背景建模与抑制的方法，其特征在于，所述方法还包括：
    循环步骤：确定下一块待分析子图像块，并通过重复所述建模步骤和所述抑制步骤来对新的待分析子图像块进行海背景抑制，直至完成所有子图像块的海背景抑制。
  3. 如权利要求1所述的高分辨遥感海洋图像的海背景建模与抑制的方法，其特征在于，所述抑制步骤具体包括：
    利用所述建模步骤构建出的所述海背景幅度谱高斯概率模型图，计算该待分析子图像块的滤波马氏距离图；
    利用所述滤波马氏距离图设计得到该待分析子图像块的理想海背景抑制滤波器；
    利用所述理想海背景抑制滤波器进一步设计高斯海背景抑制滤波器；
    利用所述高斯海背景抑制滤波器对该待分析子图像块进行频域滤波，并将滤波后的频谱图像通过二维傅里叶反变换得到经过背景抑制的子图像块。
  4. 一种高分辨遥感海洋图像的海背景建模与抑制的系统，其特征在于，所述系统包括：
    切割模块，用于对遥感图像进行预处理和分块切割，并将切割出的各个子图像块进行粗分类，选出可分析海洋图像的子图像块集合；
    分类模块，用于对可分析海洋图像的子图像块集合中的所有子图像块进一步分类为纯海水图像块子集合以及非纯海水图像块子集合；
    计算模块，用于对可分析海洋图像的子图像块集合中的每一个子图像块进行二维离散傅立叶变换，得到对应的频谱图，并由频谱图计算出对应的幅度谱图；
    建模模块，用于在可分析海洋图像的子图像块集合中确定一块待分析子图像块，以该待分析子图像块为中心，在纯海水图像块子集合中通过找出该待分析子图像块周围的所有纯海水图像块来构建海背景幅度谱高斯概率模型图；
    抑制模块，用于利用所述幅度谱高斯概率模型图构建海背景抑制滤波器，并对该待分析子图像块进行海背景抑制。
  5. 如权利要求4所述的高分辨遥感海洋图像的海背景建模与抑制的系统，其特征在于，所述系统还包括：
    循环模块，用于确定下一块待分析子图像块，并通过重复所述建模步骤和所述抑制步骤来对新的待分析子图像块进行海背景抑制，直至完成所有子图像块的海背景抑制。
  6. 如权利要求4所述的高分辨遥感海洋图像的海背景建模与抑制的系统，其特征在于，所述抑制模块具体用于：
    利用所述建模模块构建出的所述海背景幅度谱高斯概率模型图，计算该待分析子图像块的滤波马氏距离图；
    利用所述滤波马氏距离图设计得到该待分析子图像块的理想海背景抑制滤 波器；
    利用所述理想海背景抑制滤波器进一步设计高斯海背景抑制滤波器；
    利用所述高斯海背景抑制滤波器对该待分析子图像块进行频域滤波，并将滤波后的频谱图像通过二维傅里叶反变换得到经过背景抑制的子图像块。
PCT/CN2016/087716 2016-06-29 2016-06-29 高分辨遥感海洋图像的海背景建模与抑制的方法及系统 Ceased WO2018000252A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2016/087716 WO2018000252A1 (zh) 2016-06-29 2016-06-29 高分辨遥感海洋图像的海背景建模与抑制的方法及系统
US15/857,551 US10497128B2 (en) 2016-06-29 2017-12-28 Method and system for sea background modeling and suppression on high-resolution remote sensing sea images

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2016/087716 WO2018000252A1 (zh) 2016-06-29 2016-06-29 高分辨遥感海洋图像的海背景建模与抑制的方法及系统

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US15/857,551 Continuation US10497128B2 (en) 2016-06-29 2017-12-28 Method and system for sea background modeling and suppression on high-resolution remote sensing sea images

Publications (1)

Publication Number Publication Date
WO2018000252A1 true WO2018000252A1 (zh) 2018-01-04

Family

ID=60785860

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/087716 Ceased WO2018000252A1 (zh) 2016-06-29 2016-06-29 高分辨遥感海洋图像的海背景建模与抑制的方法及系统

Country Status (2)

Country Link
US (1) US10497128B2 (zh)
WO (1) WO2018000252A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110443806A (zh) * 2019-04-30 2019-11-12 浙江大学 一种基于目标增强处理的水面透明漂浮危化品图像分割方法
CN114008666A (zh) * 2019-06-18 2022-02-01 华为技术有限公司 动态图像分辨率评估
CN114782800A (zh) * 2022-02-24 2022-07-22 中国电子科技集团公司第五十四研究所 一种基于关键点回归的卫星影像舰船部件检测方法
CN115409735A (zh) * 2022-09-05 2022-11-29 航天宏图信息技术股份有限公司 一种InSAR干涉图像的滤波方法及装置

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109447089B (zh) * 2018-10-16 2021-09-14 同济大学 基于超分技术的高分辨率北极海冰类型提取方法
CN110188682B (zh) * 2019-05-30 2021-01-19 西安电子科技大学 基于几何结构双路卷积网络的光学遥感图像目标检测方法
CN110826509A (zh) * 2019-11-12 2020-02-21 云南农业大学 一种基于高分遥感影像的草原围栏信息提取系统及方法
CN111754524B (zh) * 2020-06-22 2024-01-16 厦门大学 一种基于岸线数据的遥感图像快速海陆分割方法
CN112150482B (zh) * 2020-09-17 2022-10-14 北京航空航天大学 一种基于高程信息和归一化海水指数的海陆分割方法
CN114415118B (zh) * 2021-12-30 2024-04-30 西北工业大学 一种基于二维分形特征的海面目标检测方法
CN114529520B (zh) * 2022-01-25 2024-11-26 上海卫星工程研究所 定位精度评价方法
CN116310875A (zh) * 2023-03-24 2023-06-23 中国工商银行股份有限公司 卫星遥感影像的目标检测方法及装置

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103544706A (zh) * 2013-10-28 2014-01-29 中国海洋大学 一种面向海洋数据可视化的图像序列匹配方法
CN103971370A (zh) * 2014-05-15 2014-08-06 中国科学院遥感与数字地球研究所 一种针对遥感大图像的海洋溢油智能检测方法
CN105279782A (zh) * 2015-07-02 2016-01-27 苏州蜗牛数字科技股份有限公司 实时海洋系统的模拟和渲染方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7248737B2 (en) * 2001-10-02 2007-07-24 Siemens Corporate Research, Inc. Page decomposition using local orthogonal transforms and a map optimization

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103544706A (zh) * 2013-10-28 2014-01-29 中国海洋大学 一种面向海洋数据可视化的图像序列匹配方法
CN103971370A (zh) * 2014-05-15 2014-08-06 中国科学院遥感与数字地球研究所 一种针对遥感大图像的海洋溢油智能检测方法
CN105279782A (zh) * 2015-07-02 2016-01-27 苏州蜗牛数字科技股份有限公司 实时海洋系统的模拟和渲染方法

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110443806A (zh) * 2019-04-30 2019-11-12 浙江大学 一种基于目标增强处理的水面透明漂浮危化品图像分割方法
CN110443806B (zh) * 2019-04-30 2022-05-03 浙江大学 一种基于目标增强处理的水面透明漂浮危化品图像分割方法
CN114008666A (zh) * 2019-06-18 2022-02-01 华为技术有限公司 动态图像分辨率评估
CN114782800A (zh) * 2022-02-24 2022-07-22 中国电子科技集团公司第五十四研究所 一种基于关键点回归的卫星影像舰船部件检测方法
CN115409735A (zh) * 2022-09-05 2022-11-29 航天宏图信息技术股份有限公司 一种InSAR干涉图像的滤波方法及装置

Also Published As

Publication number Publication date
US10497128B2 (en) 2019-12-03
US20180144478A1 (en) 2018-05-24

Similar Documents

Publication Publication Date Title
WO2018000252A1 (zh) 高分辨遥感海洋图像的海背景建模与抑制的方法及系统
Chen et al. Ship detection from coastal surveillance videos via an ensemble Canny-Gaussian-morphology framework
CN110310264B (zh) 一种基于dcnn的大尺度目标检测方法、装置
CN107145874B (zh) 复杂背景sar图像中的舰船目标检测与鉴别方法
CN110414411B (zh) 基于视觉显著性的海面船只候选区域检测方法
CN103400156B (zh) 基于cfar和稀疏表示的高分辨sar图像舰船检测方法
CN115983141B (zh) 一种基于深度学习反演海浪波高的方法、介质及系统
CN106384344A (zh) 一种光学遥感图像海面舰船目标检测与提取方法
CN116740528B (zh) 一种基于阴影特征的侧扫声呐图像目标检测方法及系统
CN115797813B (zh) 基于航拍图像的水环境污染检测方法
CN107808141A (zh) 一种基于深度学习的输电线路绝缘子爆裂识别方法
CN101114337A (zh) 一种地面建筑物识别定位方法
CN104240257A (zh) 基于变化检测技术的sar图像舰船目标鉴别方法
CN114821358B (zh) 光学遥感图像海上舰船目标提取与识别方法
CN112329677A (zh) 基于特征融合的遥感图像河道目标检测方法和装置
CN115294322A (zh) 水下船底可疑目标检测方法、装置、电子设备及可读介质
CN115170941A (zh) 复杂背景下声纳图像的水下小目标检测方法及系统
CN116258962B (zh) 一种结合形状置信度的宽幅sar图像低虚警目标检测方法
CN111340765B (zh) 一种基于背景分离的热红外图像倒影检测方法
CN113205139A (zh) 一种基于密度聚类的无人艇水上球体检测方法
CN120375236B (zh) 基于全方位特征融合的无人机海上目标动态识别方法
CN104881670B (zh) 一种用于sar方位角估计的快速目标提取方法
CN114140698A (zh) 一种基于FasterR-CNN的水系信息提取算法
CN118968273A (zh) 一种星上sar图像海洋内波检测方法及其相关设备
CN111695524A (zh) 一种遥感图像海面舰船检测方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16906653

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 06/05/2019)

122 Ep: pct application non-entry in european phase

Ref document number: 16906653

Country of ref document: EP

Kind code of ref document: A1