CN105551029A - Multi-spectral remote sensing image-based ship detection method - Google Patents

Abstract

The invention discloses a multi-spectral remote sensing image-based ship target quick-detection method, and belongs to the technical field of remote sensing image-based target detection. According to the method, the information of the spectral domain and the spatial domain is fully utilized, and the Walsh-Hadamard transform domain coefficient of a multi-spectral remote sensing image is processed to extract the saliency characteristic of the spatial domain. The method is used for sea ship target detection. The method of the invention can effectively overcome the defects of the traditional multi-spectral image ship detection method such as high computational complexity and complex parameter setting. A large number of real multi-spectral remote sensing data experiment results show that the method can achieve an effect of quick and accurate sea ship target detection, and is of high robustness to spectral noise and cluttered marine background. The method is of great application value in aspects like marine fisheries management, marine transportation control and military maritime monitoring.

Description

A kind of multi-spectral remote sensing image Ship Detection

Technical field

The invention belongs to Remote Sensing Target detection technique field, be specifically related to a kind of method that multi-spectral remote sensing image Ship Target detects fast.

Background technology

Multispectral imaging sensor can obtain the spectral information of each spectral coverage tie substance in spectrum dimension, obtains the spatial information of scene in space dimension simultaneously, forms the multidimensional data body containing abundant geographical environmental information.Owing to manually building target and natural material background has larger difference in the spectral characteristic of each spectral coverage, so multispectral image data have unique advantage in ground object target detects, can be used in Automatic Targets task.Especially, in maritime transportation, marine fisheries management and military monitoring etc., the marine vessel of multi-spectral remote sensing image detects and more has great significance.

Conventional multispectral object detection method is normally based on the statistically detection method of spectral information, namely by supposing that measured value is made up of background, target and noise, utilize the method difference structural setting of statistics and the model of target, recycling test of hypothesis obtains the testing result differentiating target, representative method is wherein exactly hyperchannel constant false alarm rate (constantfalsealarmrate, the CFAR) method that Reed etc. proposes.In the multispectral image of different scene, the statistical property of spectrum picture is change, and the target of CFAR method adaptively finds a detection threshold, to maintain constant detection false alarm rate in different spectrum picture object detection task.The main shortcoming of CFAR method is, if the spectrum picture signal of target in gray level with its around environment facies seemingly, so will inevitably cause false-alarm, Automatic Targets will become difficulty.

Although multispectral imaging sensor can be offered help for the man-made target in mixed and disorderly background detects, two difficulties will be faced in the practical application that automatic naval vessel detects.First, sea wind, ocean current, the tail of ship, the existence of the situations such as leakage of oil, will cause the change of seawater radiation or spectral reflection characteristic.The spectral reflection characteristic on middle naval vessel of advancing also is unstable, and is difficult to estimate.All these situations all can cause wave clutter (sea clutter) and the Ship Target overlap in the statistical distribution of each spectral coverage emittance.Secondly, marine monitoring system needs one algorithm of target detection fast, and need real-time analysis due to it and process a large amount of multispectral datas, this requirement is also for automatic Target Detection is challenged.In order to distinguish interesting target from false-alarm, classic method needs the algorithm comprising target identification usually, and this identifying is usually too complicated for computing real-time system.In addition, in order to judge the existence of target, classic method often needs to verify all image-regions, but be in fact concerned about content only accounts for a very little part in image usually.This comprehensive processing process both can cause and calculate waste, had increased the weight of again analysis difficulty.

Summary of the invention

The object of the invention is to the multi-spectral remote sensing image Ship Detection proposing a kind of view-based access control model conspicuousness mechanism, its computation complexity is low, optimum configurations is simple, the Ship Target that can accurately and effectively detect in multi-spectral remote sensing image.

For reaching above-mentioned purpose of the present invention, multi-spectral remote sensing image Ship Detection provided by the invention, specifically comprises the following steps:

1) by spatial resolution be M × N pixel multi-spectral remote sensing image l tie up spectroscopic data be considered as l width gray level image X _i, i=1 ..., l, carries out Walsh Hadamard transform to every one dimension spectrum picture respectively;

2) in Walsh Hadamard transform territory, operation is normalized to the Walsh Hadamard transform domain coefficient of every one dimension spectrum picture, is set to 1 by all on the occasion of element, all negative value elements are set to-1;

3) carry out the inverse transformation of corresponding Walsh Hadamard to every one dimension normalized Walsh Hadamard transform domain coefficient, the spectral coverage calculating every one dimension spectrum picture is significantly schemed;

4) the remarkable figure of spectral coverage of all dimension spectrum pictures is sued for peace on Spatial Dimension, then carry out slight smoothing processing by the result of Gaussian filter to summation, calculate final synthesis and significantly scheme;

5) detection threshold of Ship Target is gone out according to the mean value computation of final significantly figure;

6) utilize the automatic detection of the detection threshold of Ship Target realization to Ship Target, obtain the testing result of binaryzation.

Wherein, in a first step, the specific formula for calculation of Walsh Hadamard transform carries out to input multispectral image as follows:

F _i＝HX _iW ^T,i＝1,...,l

Wherein, H represents M rank hadamard matrix, and W represents N rank hadamard matrix, F _irepresent the i-th dimension spectrum picture X _iwalsh Hadamard transform domain coefficient matrix.

Simultaneously, the spatial resolution of the image of input is M × N, and M and N meets the requirement of the sequence length of Walsh Hadamard transform, if do not meet, the mode of operation of entire image being carried out to piecemeal process can be taked, in order to retain small and weak ship signaling, down-sampled process can not be carried out to the original image of input.

Wherein, in second step, specific formula for calculation Walsh Hadamard transform domain coefficient being normalized to operation is:

$B_i(x,y)=\frac{F_i(x,y)}{\left|F_i(x,y)\right|},i=1,...,l$

Wherein, || represent the symbol that takes absolute value, B _irepresent the i-th dimension spectrum picture X _inormalized Walsh Hadamard transform domain coefficient matrix.

Wherein, in the third step, the specific formula for calculation of the remarkable figure of spectral coverage of every one dimension spectrum picture is as follows:

S _i＝abs(H ^TB _iW),i＝1,...,l

Wherein, abs () represents and to take absolute value operation to each element of input matrix.

Wherein, in the 4th step, the specific formula for calculation of the remarkable figure of final synthesis is:

$\tilde{S}=S_1+S_2+...+S_l,S=G*\tilde{S}$

Wherein, G represents the gaussian kernel of 2 dimensions, and S represents that the synthesis finally obtained significantly is schemed.

The present invention is in the computation process of the remarkable figure of spectral coverage of every one dimension spectrum picture, high frequency noise (as sea clutter noise) also can by violent amplification, but these high frequency noises be exaggerated are incoherent in each spectral coverage, and the ship signaling in the remarkable figure of each spectral coverage is relevant.Therefore, each spectral coverage is significantly schemed be added, these incoherent high frequency noises in the remarkable figure of each spectral coverage can be allowed to cancel out each other, mutually suppress, but also the ship signaling of the statistical correlation in the remarkable figure of each spectral coverage can be allowed mutually to superpose further strengthened.However, the result after all spectral coverages significantly scheme summation in still can remain some high frequency noises do not curbed, the result that the existence of these noises can detect to naval vessel brings false-alarm, in order to eliminate the detection false-alarm that these residual high frequency noises may bring, the result that the present invention uses a gaussian kernel G with suitable parameters significantly to scheme to be added to all spectral coverages carry out low pass smothing filtering, to allow ship signaling and ambient noise signal have obvious grey scale contrasts, ship signaling can also be retained well simultaneously.

Wherein, in the 5th step, the specific formula for calculation of the detection threshold of Ship Target is:

$\mathrm{\θ}=\frac{\mathrm{\α}}{M\×N}\underset{x=0}{\overset{M-1}{\Σ}}\underset{y=0}{\overset{N-1}{\Σ}}S(x,y)$

Wherein, θ is the detection threshold of Ship Target, M and N is the length of remarkable figure and wide, and α is the empirical value obtained from the multispectral image data naval vessel test experience of the different scene of many groups, through great many of experiments, find that arranging 3≤α≤5 can obtain good testing result.

When multispectral image exists Ship Target, the Ship Target in the remarkable figure of synthesis is highlighted, and its saliency value is larger, and marine background region is relatively dark, and its saliency value is less.In this case, the saliency value of Ship Target can be more much bigger than the average of the remarkable figure of synthesis, and be greater than the detection threshold obtained by remarkable figure mean value computation, so Ship Target can be detected accurately.When multispectral image does not have Ship Target, the highlighted target of giving prominence to especially would not be there is in the remarkable figure of synthesis.In this case, the maximal value of synthesizing remarkable figure can not be more much larger than the mean value of the remarkable figure of synthesis, and the maximal value of remarkable figure also can be less than the detection threshold obtained by the remarkable figure mean value computation of synthesis, thus avoid the generation of false-alarm.

Wherein, in the 6th step, to the specific formula for calculation that Ship Target detects automatically be:

$D(x,y)=\left\{\begin{array}{cc}1,& S(x,y)\&GreaterEqual;\mathrm{\&theta;}\\ 0,& S(x,y)<\mathrm{\&theta;}\end{array}\right.$

Wherein, D is binaryzation testing result, and image-region judgement detected value being equaled 1 is Ship Target, and image-region judgement detected value being equaled 0 is marine background.

Multi-spectral remote sensing image Ship Target Detection method proposed by the invention is the method for view-based access control model conspicuousness mechanism, and the method utilizes the Walsh Hadamard transform domain coefficient of multispectral image data to obtain significantly figure for Ship Target Detection.Compared with traditional multispectral object detection method, the present invention does not need the prerequisite using some Utopian hypothesis of the probability density characteristics to Sea background and target as modeling; It does not rely on priori, does not have the optimum configurations of many complexity yet.Simultaneously, because Walsh Hadamard transform exists quick implementation algorithm, method in the present invention can calculate the remarkable figure of the multispectral data of input fast, the requirement of process in real time in practical application can be met, for multi-spectral remote sensing image Ship Target Detection provides a kind of effective fast algorithm newly.

Accompanying drawing explanation

Fig. 1 is the multi-spectral remote sensing image Ship Detection process flow diagram of the invention process row;

Fig. 2 is the present invention to the naval vessel detection example of multi-spectral remote sensing image having Ship Target;

Wherein: 6 spectral coverages of the true Landsat7 remotely-sensed data of (a)-(f) East China Sea subregion; G synthesis that () calculates significantly is schemed; (h) testing result;

Fig. 3 is the naval vessel detection example of the present invention to the multi-spectral remote sensing image without Ship Target;

Wherein: 6 spectral coverages of the true Landsat7 remotely-sensed data of (a)-(f) East China Sea subregion; G synthesis that () calculates significantly is schemed; (h) testing result;

Embodiment

Below by example, the present invention will be further described.It should be noted that the object publicizing and implementing example is to help to understand the present invention further, but it will be appreciated by those skilled in the art that: in the spirit and scope not departing from the present invention and claims, various substitutions and modifications are all possible.Therefore, the present invention should not be limited to the content disclosed in embodiment, and the scope that the scope of protection of present invention defines with claims is as the criterion.

Fig. 1 is the processing flow chart of multi-spectral remote sensing image Ship Detection of the present invention, comprising:

The first step, carries out Walsh Hadamard transform by every one dimension spectrum picture of multispectral data

By spatial resolution be the multi-spectral remote sensing image of M × N pixel l tie up spectroscopic data be considered as l width gray level image X _i, i=1 ..., l, carry out Walsh Hadamard transform to every one dimension spectrum picture respectively, specific formula for calculation is as follows:

F _i＝HX _iW ^T,i＝1,...,l

Wherein, H represents M rank hadamard matrix, and W represents N rank hadamard matrix, F _irepresent the i-th dimension spectrum picture X _iwalsh Hadamard transform domain coefficient matrix.

Second step, is normalized operation to the Walsh Hadamard transform domain coefficient of every one dimension spectrum picture

In Walsh Hadamard transform territory, be normalized operation to the Walsh Hadamard transform domain coefficient of every one dimension spectrum picture, be set to 1 by all on the occasion of element, all negative value elements are set to-1, and specific formula for calculation is as follows:

$B_i(x,y)=\frac{F_i(x,y)}{\left|F_i(x,y)\right|},i=1,...,l$

Wherein, || represent the symbol that takes absolute value, B _irepresent the i-th dimension spectrum picture X _inormalized Walsh Hadamard transform domain coefficient matrix.

3rd step, does the inverse transformation of Walsh Hadamard to every one dimension normalization coefficient, and the spectral coverage calculating every one dimension spectrum picture is significantly schemed

Carry out the inverse transformation of corresponding Walsh Hadamard to every one dimension normalized Walsh Hadamard transform domain coefficient, the spectral coverage calculating every one dimension spectrum picture is significantly schemed, and specific formula for calculation is as follows:

S _i＝abs(H ^TB _iW),i＝1,...,l

Wherein, abs () represents and to take absolute value operation to each element of input matrix.

4th step, significantly schemes addition and obtains a width synthesis significantly figure by all spectral coverages

Sued for peace on Spatial Dimension by the remarkable figure of spectral coverage of all dimension spectrum pictures, calculate final synthesis and significantly scheme, specific formula for calculation is as follows:

$\tilde{S}=S_1+S_2+...+S_l,S=G*\tilde{S}$

Wherein, G represents the gaussian kernel of 2 dimensions, and S represents that the synthesis finally obtained significantly is schemed.In this step, the present invention uses one to be of a size of 3 × 3 (or 5 × 5), and standard deviation is (0.5 ~ 1) of 0.75, and rotational symmetric gauss low frequency filter is to the result be added carry out slight smoothing processing.

5th step, goes out the detection threshold of Ship Target according to the remarkable figure mean value computation of synthesis

Go out the detection threshold of Ship Target according to the mean value computation of final significantly figure, specific formula for calculation is as follows:

$\mathrm{\&theta;}=\frac{\mathrm{\&alpha;}}{M\&times;N}\underset{x=0}{\overset{M-1}{\&Sigma;}}\underset{y=0}{\overset{N-1}{\&Sigma;}}S(x,y)$

Wherein, θ is the detection threshold of Ship Target, M and N is the length of remarkable figure and wide, α is the empirical value obtained from the multispectral image data naval vessel test experience of the different scene of many groups, through great many of experiments, discovery arranges 3≤α≤5 can obtain good testing result, as got the remarkable figure average of α=4 times as detection threshold, this threshold value can be less than the saliency value of Ship Target in the remarkable figure of synthesis when there being Ship Target, and the maximum saliency value of the remarkable figure of synthesis can be greater than when there is no Ship Target, that is, it both accurately can detect Ship Target in the scene having Ship Target, detection false-alarm can be effectively prevent again in the scene not having Ship Target.

6th step, utilizes the automatic detection of detection threshold realization to Ship Target

Utilize the automatic detection of the detection threshold of Ship Target realization to Ship Target, obtain the testing result of binaryzation, specific formula for calculation is as follows:

$D(x,y)=\left\{\begin{array}{cc}1,& S(x,y)\&GreaterEqual;\mathrm{\&theta;}\\ 0,& S(x,y)<\mathrm{\&theta;}\end{array}\right.$

Wherein, D is binaryzation testing result, the region representation Ship Target that value equals 1, the region representation marine background that value equals 0.

Be illustrated in figure 2 the example that above-mentioned processing procedure process one scape has the multi-spectral remote sensing image of Ship Target.Fig. 2 (a)-(f) is depicted as the image of 6 spectral coverages of the true Landsat7 remotely-sensed data of East China Sea subregion, wherein contains some Ship Target signals.In the remarkable figure of synthesis that this method shown in Fig. 2 (g) calculates, remarkable figure is full resolution, and Ship Target obtains and to strengthen and from complex background saliency out.Be depicted as at Fig. 2 (h) the Ship Target Detection result that this method calculates, white portion represents the Ship Target detected, black region represents marine background.

Be illustrated in figure 3 the example that above-mentioned processing procedure process one scape does not have the multi-spectral remote sensing image of Ship Target.Fig. 3 (a)-(f) is depicted as the image of 6 spectral coverages of the true Landsat7 remotely-sensed data of East China Sea subregion, wherein without any Ship Target signal.In the remarkable figure of synthesis that this method shown in Fig. 3 (g) calculates, due to without any Ship Target, so not by the marking area strengthened especially in remarkable figure.In the Ship Target Detection result that this method shown in Fig. 3 (h) calculates, region-widely represent that detection does not cause any false-alarm for black.

Multi-spectral remote sensing image Ship Detection disclosed by the invention, only utilize the coefficient of the Walsh Hadamard transform of each dimension spectrum picture of multispectral data to calculate the conspicuousness of corresponding scene, to the normalization operation that coefficient in transform domain adopts, there is biorational, it simulates the process that in human brain primary visual cortex, homogenous characteristics suppresses mutually, can obtain full resolution remarkable figure and by Ship Target from marine background saliency out, utilize ground unrest in the remarkable figure of each spectral coverage uncorrelated and this characteristic of ship signaling statistical correlation, remarkable for each spectral coverage figure is carried out suing for peace on Spatial Dimension thus effectively weakens and inhibit the high frequency noise in remarkable figure, the detection threshold utilizing the remarkable figure average of synthesis to determine both can detect Ship Target accurately in the scene having Ship Target, effectively can avoid again detecting false-alarm in the scene not having Ship Target, method is simple, efficiently, accurately.The present invention achieves the result being obviously better than other classic methods in a large amount of multispectral data test.

Although the present invention discloses as above with preferred embodiment, but and be not used to limit the present invention.Any those of ordinary skill in the art, do not departing under technical solution of the present invention ambit, the Method and Technology content of above-mentioned announcement all can be utilized to make many possible variations and modification to technical solution of the present invention, or be revised as the Equivalent embodiments of equivalent variations.Therefore, every content not departing from technical solution of the present invention, according to technical spirit of the present invention to any simple modification made for any of the above embodiments, equivalent variations and modification, all still belongs in the scope of technical solution of the present invention protection.

Claims (8)

1. a multi-spectral image ship detection method, is characterized in that, comprises the following steps:

1) by spatial resolution be M × N pixel multi-spectral remote sensing image l tie up spectroscopic data be considered as l width gray level image X _i, i=1 ..., l, carries out Walsh Hadamard transform to every one dimension spectrum picture respectively;

2) in Walsh Hadamard transform territory, operation is normalized to the Walsh Hadamard transform domain coefficient of every one dimension spectrum picture, is set to 1 by all on the occasion of element, all negative value elements are set to-1;

3) carry out the inverse transformation of corresponding Walsh Hadamard to every one dimension normalized Walsh Hadamard transform domain coefficient, the spectral coverage calculating every one dimension spectrum picture is significantly schemed;

4) the remarkable figure of spectral coverage of all dimension spectrum pictures is sued for peace on Spatial Dimension, then with the result smoothing process of Gaussian filter to summation, calculate final synthesis and significantly scheme;

5) detection threshold of Ship Target is gone out according to the mean value computation of final significantly figure;

6) utilize the automatic detection of the detection threshold of Ship Target realization to Ship Target, obtain the testing result of binaryzation.

2. multi-spectral image ship detection method as claimed in claim 1, is characterized in that, step 1) in Walsh Hadamard transform specific formula for calculation as follows:

F _i＝HX _iW ^T,i＝1,...,l

Wherein, H represents M rank hadamard matrix, and W represents N rank hadamard matrix, F _irepresent the i-th dimension spectrum picture X _iwalsh Hadamard transform domain coefficient matrix.

3. multi-spectral image ship detection method as claimed in claim 1, it is characterized in that, step 1) in the spatial resolution of image be M × N, wherein M and N meets the requirement of the sequence length of Walsh Hadamard transform, if do not meet, take to carry out piecemeal process to entire image.

4. multi-spectral image ship detection method as claimed in claim 1, is characterized in that, step 2) in the specific formula for calculation that Walsh Hadamard transform domain coefficient is normalized operation be:

$B_i(x,y)=\frac{F_i(x,y)}{\left|F_i(x,y)\right|},i=1,...,l$

Wherein, || represent the symbol that takes absolute value, B _irepresent the i-th dimension spectrum picture X _inormalized Walsh Hadamard transform domain coefficient matrix.

5. multi-spectral image ship detection method as claimed in claim 1, is characterized in that, step 3) in the specific formula for calculation of the remarkable figure of spectral coverage of every one dimension spectrum picture as follows:

S _i＝abs(H ^TB _iW),i＝1,...,l

Wherein, abs () represents and to take absolute value operation to each element of input matrix.

6. multi-spectral image ship detection method as claimed in claim 1, is characterized in that, step 4) in the specific formula for calculation of the final remarkable figure of synthesis be:

$\tilde{S}=S_1+S_2+...+S_l,S=G*\tilde{S}$

Wherein, G represents the gaussian kernel of 2 dimensions, and S represents that the synthesis finally obtained significantly is schemed.

7. multi-spectral image ship detection method as claimed in claim 1, is characterized in that, step 5) in the specific formula for calculation of detection threshold of Ship Target be:

$\mathrm{\&theta;}=\frac{\mathrm{\&alpha;}}{M\&times;N}\underset{x=0}{\overset{M-1}{\&Sigma;}}\underset{y=0}{\overset{N-1}{\&Sigma;}}S(x,y)$

Wherein, θ is the detection threshold of Ship Target, M and N is the length of remarkable figure and wide, and α is the empirical value obtained from the multispectral image data naval vessel test experience of the different scene of many groups, 3≤α≤5.

8. multi-spectral image ship detection method as claimed in claim 1, is characterized in that, step 6) in the specific formula for calculation that Ship Target detects automatically be:

$D(x,y)=\left\{\begin{array}{cc}1,& S(x,y)\&GreaterEqual;\mathrm{\&theta;}\\ 0,& S(x,y)<\mathrm{\&theta;}\end{array}\right.$

Wherein, D is binaryzation testing result, the region representation Ship Target that value equals 1, the region representation marine background that value equals 0.