CN101414009B - Portable airborne offshore oil overflowing remote sense monitoring system - Google Patents

Abstract

The invention provides a portable airborne marine oil spill remote sensing monitoring system, and relates to the marine pollution monitoring field. The system comprises an infrared/ultraviolet/multispectral digital camera, a central control and display system as well as a data management and image processing operation software system. The central control and display system records image information acquired by the infrared/ultraviolet/multispectral digital camera, and superposes the information with an electronic chart for on-line or off-line track playback; and the data management and image processing operation software system which is arranged in a portable industrial control computer of the central control and display system reads the image information, identifies oil spill on line, computes contaminated area on line and identifies thick oil film. The system has the advantages that the portable airborne marine oil spill remote sensing monitoring system can be expediently equipped on various ocean patrol planes without power supply from the planes; when a marine oil spill accident happens, the system can track the marine oil spill in real time, estimate oil spillage and the contaminated area, and evaluate the oil spill elimination effect.

Description

Portable airborne offshore oil overflowing remote sense monitoring system

Technical field

The present invention relates to the marine pollution monitoring technical field, relate in particular to a kind of portable airborne offshore oil overflowing remote sense monitoring system.

Background technology

The airborne offshore oil overflowing remote sense monitoring system product has the AIREYE (infrared/visible light) of the U.S., the marine surveillance monitoring system MSS of Sweden etc., the foremost MSS5000 that will belong to Sweden aerospace company in the world.At present, cured, Dutch, the Finland in Spain, Sweden, west, the U.S. etc. much developed countries all equipped MSS5000.

Abroad have: Airborne IR/ultraviolet oil overflowing remote sense detector (U.S.), the patent No.: 5461236 in the patent aspect the airborne offshore oil overflowing remote sense monitoring; Monitoring and measurement oil spilling method and instrument (U.S.), the patent No.: 7009550.A large amount of documents shows that external airborne offshore oil overflowing remote sense monitoring technology is made up of the two large divisions: airborne oil overflowing remote sense detector (hardware) and monitoring of airborne oil overflowing remote sense and measuring method (algorithm and software).The Real time identification offshore spilled oil and distinguish oil film definitely and this difficult problem of doubtful oil film on all adopt airborne many remote sensors integrated technology and many remote sensors information fusion technology abroad.As Sweden MSS5000 system is that multinomial remote sensing technologies such as infrared, ultraviolet, visible light, side-looking radar and laser-induced fluorescence (LIF) integrate.In a word, external airborne offshore oil overflowing remote sense monitoring system integration is strong, can realize round-the-clock full meteorology, marine oil overflow active monitoring on a large scale.External mobile system product is the software and hardware bundle sale.External airborne offshore oil overflowing monitoring system products-hardware is more advanced, but the weak point of software aspect is a little less than the ability of real-time information processing, particularly Real time identification marine oil overflow, thicker oil film and the calculating contaminated area aspect of identification.

Summary of the invention

The purpose of this invention is to provide a kind of portable airborne offshore oil overflowing remote sense monitoring system, can the Real time identification marine oil overflow, the thicker oil film of identification and calculate the contaminated area aspect.

In order to achieve the above object, technical scheme of the present invention is as follows:

Portable airborne offshore oil overflowing remote sense monitoring system of the present invention mainly is made up of infrared/ultraviolet/multispectral digital camera, central control and display system, data management and image processing operations software systems, described infrared/ultraviolet/multispectral digital camera is used for obtaining respectively the infrared/ultraviolet/multispectral digital image information on monitored sea; Control of described central authorities and display system are by portable industrial pc, GPS, GIS (Geographic Information System) and system of image acquisition function software form, be used to realize that offshore oil overflowing remote sense surveillance map picture shows, offshore oil overflowing remote sense surveillance map picture (multispectral digital camera image, infrared image, ultraviolet image) storage, the stack of marine oil overflow monitoring remote sensing images and electronic chart, the playback of marine oil overflow monitoring remote sensing images, the daily record of cruising, the management of oil spilling image file, the time recording of flight path, individual collection and the continuous acquisition of single-sensor, the collection in turn of multisensor, the record of video information, online flight path playback; Described data management and image processing operations software systems are installed in the portable industrial pc of central authorities control and display system, be used to read offshore oil overflowing remote sense monitoring image information and finish the ONLINE RECOGNITION oil spilling, at line computation contaminated area and the thick oil film of ONLINE RECOGNITION.

The present invention adopts infrared remote sensing, ultraviolet remote sensing, multispectral remote sensing monitoring marine oil overflow, and its principle is as follows:

Infrared remote sensing: utilize the heat radiation feature difference between oil spilling and the seawater, the bright temperature gray-scale value on infrared image has tangible difference, through Digital Image Processing, enlarges the difference of these environment parameters and seawater, reaches the purpose of monitoring.The shortcoming of infrared remote sensing technology is to distinguish hydrophyte, oil spilling, sea ice.But the price of infrared remote sensor is more and more lower, and its weight is also more and more lighter, need not aircraft is transformed, and is that maximum marine oil spill prospecting tools is adopted in present countries in the world.

Ultraviolet remote sensing: compare with seawater, because oil film is very high to the reflection of UV radiation, therefore can utilize the imaging of ultraviolet remote sensing device to distinguish oil spilling and seawater, even oil film thickness also can utilize the imaging of ultraviolet remote sensing device to distinguish oil film and seawater less than 0.05 μ m.The shortcoming of ultraviolet remote sensing device is that it can not distinguish hydrophyte, sea grass under water and oil spilling waterborne.

Multispectral remote sensing: with the organism of seawater, red tide, sea ice, dissolving, seabeach sand, stone, sewage, waterborne (in) to compare oil film be at wavelength 420nm, 480nm and 520nm to the reflection sensitive band of sunlight to plant etc.Because it is significantly different that other material and the reflection sensitive band of oil spilling have, so its confirmable differentiation oil spilling and other material.

The invention has the beneficial effects as follows: this portable airborne offshore oil overflowing remote sense monitoring system can be equipped on the various ocean patrol aircrafts easily, does not need aircraft that power supply is provided; When the marine oil overflow accident takes place, this system's energy real-time follow-up marine oil overflow, estimation oil spillage and contaminated area are also removed effect to oil spilling and are assessed.

Description of drawings

Fig. 1 is PCA﹠amp of the present invention; The structural representation of BP method modeling.

Fig. 2 is a remote sensing monitoring oil spill area intelligent algorithm process flow diagram of the present invention.

Embodiment

Below in conjunction with accompanying drawing the present invention is done description in further detail:

Portable airborne offshore oil overflowing remote sense monitoring system of the present invention adopts PCA﹠amp; The identification of marine oil overflow is carried out in BP (principal component analysis combines with the BP network) method modeling, and its recognizer is as follows:

1, PCA is to the identification modeling of oil spilling

1.1, the PCA algorithm

The main process of principal component analysis is divided into feature selection process and characteristic extraction procedure.In feature selection process, mainly the input space is mapped to output region, thereby obtains the feature of input, i.e. Shu Ru principal component; And characteristic extraction procedure dimensionality reduction process is just chosen main feature and is cast out the further feature component in this process.

(1) feature selection process: the purpose of this process is the mapping that realizes from the input space to the feature space, is that the feature extraction in second step lays the first stone.Therefore, the key of this process is exactly the selected characteristic vector and obtains the projection of input on all features, make these projections can satisfy the requirement of square error minimum in feature extraction, and these projections is exactly the principal component of input.

For given known probability distribution function { the m dimension random vector X=[x of X} ₁, x ₂, Λ, x _m] ^T, establish its average E[X] and be 0.(when the average of X is non-vanishing, can make x '=X-E[X], and obtain

And covariance C _XFor

Because E[X]=0, so covariance matrix is autocorrelation matrix:

Calculate C _XEigenvalue ₁, λ ₂, Λ, λ _mNormalization proper vector U with correspondence ₁, U ₂, Λ, U _mC _XU _i=λ _iU _iI=1,2, Λ, m (3)

Wherein, U _i=[u _I1, u _I2, Λ, u _In] ^TProve proper vector U in theory ₁, U ₂, Λ, U _mThe feature of the input that satisfies condition exactly.Might as well establish eigenvalue ₁〉=λ ₂〉=Λ λ _m, then $y_i=U_i^{T}X(i=\mathrm{1,2},\mathrm{\Λ\; m})$ Import projection exactly, be i the component of X proper vector.As follows with matrix representation:

Y＝U ^TX (4)

Wherein, eigenvectors matrix U=[U ₁, U ₂, Λ U _m] satisfy UU ^T=E, and Y=[y ₁, y ₂, Λ, y _m] ^TTherefore, by the n-dimensional vector Y of eigenvectors matrix U with input, a certain component y of Y _iI principal component for input X.

According to Y=U ^TX can reconstruct obtain X with the linear combination of proper vector, as shown in the formula

$X=\mathrm{UY}=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{U}_i{y}_i---\left(5\right)$

Just can recover original input vector X, finish the conversion from the feature space to the input space according to all principal components.

(2) characteristic extraction procedure: obtained all principal components by feature selecting, and in characteristic extraction procedure, then choose main feature and give up less important, thereby reach the purpose of dimensionality reduction.For this analyzes average to vectorial Y, to the average E[Y of Y], have:

Therefore, auto-covariance matrix C _YThe autocorrelation matrix of Y just:

Because U is the eigenvectors matrix of X, so have

$C_Y=\left[\begin{array}{c}{\mathrm{\λ}}_1,0,\mathrm{\Λ\Λ},0\\ 0,{\mathrm{\λ}}_2,\mathrm{\Λ\Λ},0\\ M\\ \mathrm{0,0},\mathrm{\Λ\Λ},{\mathrm{\λ}}_m\end{array}\right]---\left(8\right)$

Can find out auto-covariance matrix C by following formula _YDepend on the auto-covariance matrix of importing X.When Y was blocked, it was optimum under the square error meaning guaranteeing to block, can be according to the pairing eigenvalue of proper vector of decision principal component ₁, λ ₂, Λ, λ _m(λ ₁〉=λ ₂〉=Λ λ _m) block.The pairing eigenwert of proper vector is big more, and its contribution when reconstruct is just more little.So the proper vector very little to those eigenwerts can be ignored.Consider λ ₁, λ ₂, Λ, λ _mIn before L (eigenwert that 1≤L≤m) is maximum when the principal component that X determines with these proper vectors is come reconstruct, satisfies minimum mean square error criterion, and the estimated value of reconstruct is

$\hat{X}=\underset{i=1}{\overset{L}{\mathrm{\Σ}}}{U}_i{y}_i---\left(9\right)$

Its mean square deviation

$e_L=E\left[{(X-\hat{X})}^2\right]=\underset{i=L+1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i---\left(10\right)$

As can be seen from the above equation, work as C _XEach eigenvalue of maximum of preceding L when big, it is minimum that mean square deviation reaches.Have according to matrix theory $\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\λ}}_i=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{c}_{\mathrm{ii}},$ C in the formula _IiBe C _XThe diagonal of a matrix element, the definition variance contribution ratio:

When variance contribution ratio enough big (as reach 70%, 80% or 90% etc.), can be with preceding L proper vector u ₁, u ₂, Λ, u _LThe space that constitutes as the low-dimensional projector space (L＜＜m), thereby finish the processing of dimensionality reduction.

1.2, the identifying of PCA algorithm:

Data are input in the principal component analysis network, and output vector is set up planimetric coordinates first and second principal components as horizontal ordinate and ordinate, first and second principal components have been represented 79.765% and 9.0007% in whole one group of data characteristics respectively, so different oil spilling samples is in occupation of the zones of different on plane, thereby reached the effect of Classification and Identification.Compare with the recognition methods of more point-to-point two different glossy spectrums, the principal component analysis neural net method has the ability of one group of data sample of more powerful identification.It is a kind of recognition methods of not having the supervision pattern, needs translation.So should utilize the oil spilling sample of known kind to indicate each zone on the plane when discerning, the sample of very close known oil spilling sample just can be considered the oil spilling sample of same kind in the plane.

In principal component analysis neural network method, the spectroscopic data of input has been played the part of an extremely important role for efficient.If the data of input also are not suitable for dividing good type, whole computation process must be carried out once again.On the other hand, the scope of spectrum in characteristic relation figure that we must at first clear and definite each known oil, the position that is in of object observing spectrum then, thus judge its kind.

2, BP neural net model establishing

2.1, sample set is defined as follows in the BP network:

Input vector Ib, output vector Ob, target vector are Tb.

Wherein, input vector I is defined as the spectrum samples training data that collects.Target vector Tb is defined as:

Tb＝[1?2?3?4?5?6?7?8]；

The target substance and the theoretical output valve table of comparisons of table 1:BP network

Target substance	Acetone	Diesel oil	Machine oil	Alcohol	Gasoline	Lubricating oil	Salad oil	Seawater
									Theoretical output valve	1	2	3	4	5	6	7	8

If real output value is 0.998 ≈ 1, judge that then this kind material is an acetone; If real output value is 6.027 ≈ 6, judge that then this kind material is a lubricating oil.

For each group desired value (as alcohol 4),, this desired value 4 among 1 corresponding Tb of spectrum samples vector should be arranged in input vector P then if definition has 14 in Tb; If definition has 54 in Tb, this desired value 4 among 5 corresponding Tb of spectrum samples vector should be arranged also in input vector P then.Vector P and Tb concern that one to one the desired value among the Tb may repeatedly repeat certainly, though facts have proved the task amount that has increased network training like this, can improve the performance of network and the accuracy of classification greatly.

2.2, BP network structure design:

Input layer:

Input layer is an X neuron, and the input data I is the matrix of X * n, and column vector is corresponding to each input training sample vector, and the number of spectrum samples is trained in the n representative.

Output layer:

The output layer of BP network is made as 1 neuron, and its output valve is a value among the 1-8, so just can corresponding several materials of setting in advance.

Hidden layer:

Because be the dull nonlinear function that rises between the input and output of the implicit unit of B-P network, its output is a soft function, therefore the implicit unit number that requires than perceptron is lacked.But the number of hidden layer neuron does not have good analytic expression too.If implicit number of unit very little, system performance is undesirable; Implicit number of unit is too big, not only learning time long, system's bulky complex that becomes, and error is also not necessarily best.Therefore certainly exist the implicit unit number an of the best.In order to find this optimal number, use identical input, output neuron number, different hidden neuron numbers, the identical spectra training sample is tested.With reference to the experience choosing method, i.e. the arithmetic mean or the geometric mean of the desirable input and output interstitial content of the initial number of hidden layer neuron.

2.3, the learning training algorithm of BP network:

The 1st step was provided with variable and parameter:

X _k=[x _K1, x _K2, Λ, x _KM], (k=1,2, Λ N) is the oil spilling spectrum vector or the title training spectrum samples of input, N is the number of training spectrum samples.

$W_{\mathrm{MI}}\left(n\right)=\left[\begin{array}{c}{w}_{11}\left(n\right),w_{12}\left(n\right),\mathrm{\Λ}{w}_{1I}\left(n\right)\\ w_{21}\left(n\right),w_{22}\left(n\right),\mathrm{\Λ}{w}_{2I}\left(n\right)\\ M\\ w_{M1}\left(n\right),w_{M2}\left(n\right)\mathrm{\Λ}{w}_{\mathrm{MI}}\left(n\right)\end{array}\right]|$ Weight vector when being the n time iteration between input layer and the hidden layer I.

$W_{\mathrm{IJ}}\left(n\right)=\left[\begin{array}{c}{w}_{11}\left(n\right),w_{12}\left(n\right),\mathrm{\Λ}{w}_{1J}\left(n\right)\\ w_{21}\left(n\right),w_{22}\left(n\right),\mathrm{\Λ}{w}_{2J}\left(n\right)\\ M\\ w_{I1}\left(n\right),w_{I2}\left(n\right)\mathrm{\Λ}{w}_{\mathrm{IJ}}\left(n\right)\end{array}\right]$ Weight vector when being the n time iteration between hidden layer I and the hidden layer J.

$W_{\mathrm{JP}}\left(n\right)=\left[\begin{array}{c}{w}_{11}\left(n\right),w_{12}\left(n\right),\mathrm{\Λ}{w}_{1P}\left(n\right)\\ w_{21}\left(n\right),w_{22}\left(n\right),\mathrm{\Λ}{w}_{2P}\left(n\right)\\ M\\ w_{J1}\left(n\right),w_{J2}\left(n\right)\mathrm{\Λ}{w}_{\mathrm{JP}}\left(n\right)\end{array}\right]$ Weight vector when being the n time iteration between hidden layer J and the output layer.

Y _k(n)=[y _K1(n), y _K2(n), Λ, y _KP(n)], (k=1,2, Λ, the actual output of network when N) being the n time iteration.

d _k=[d _K1, d _K2, Λ, d _KP], (k=1,2, Λ, N) one-dimensional vector for setting, element is represented the kind of different oil, and it is the output of expectation.η is a learning rate; N is an iterations.

The 2nd step initialization is composed and is given W _MI(0), W _IJ(0), W _JP(0) each small random nonzero value.

Spectrum samples X of the 3rd step input _k, n=0.

The 4th step is to the spectrum samples X of input _k, every layer of neuronic input signal u of forward calculation BP network and output signal v.Wherein

$v_p^P\left(n\right)=y_{\mathrm{kp}}\left(n\right),$ p＝1，2，Λ，P (12)

The 5th step is by desired output d _kThe actual output Y that tries to achieve with previous step _k(n) error of calculation E (n) judges whether to meet the demands, and goes to for the 8th step if satisfy, and does not satisfy going to for the 6th step.

Whether the 6th step judged n+1 greater than maximum iteration time, if greater than going to for the 8th step, if be not more than, to input sample X _k, every layer of neuronic partial gradient δ of backwards calculation.Wherein

${\mathrm{\δ}}_p^P\left(n\right)=y_p\left(n\right)(1-y_p\left(n\right))(d_p\left(n\right)-y_p\left(n\right)),$ p＝1，2，Λ，P (13)

${\mathrm{\δ}}_j^J\left(n\right)=f\′\left(u_j^J\left(n\right)\right)\underset{p=1}{\overset{P}{\mathrm{\Σ}}}{\mathrm{\δ}}_p^P\left(n\right)w_{\mathrm{jp}}\left(n\right),$ j＝1，2，Λ，J (14)

${\mathrm{\δ}}_i^I\left(n\right)=f\′\left(u_i^I\left(n\right)\right)\underset{j=1}{\overset{J}{\mathrm{\Σ}}}{\mathrm{\δ}}_j^J\left(n\right)w_{\mathrm{ij}}\left(n\right),$ i＝1，2，Λ，I (15)

The 7th step was calculated as follows weights correction amount w, and revised weights; N=n+1 went to for the 4th step

$\mathrm{\Δ}{w}_{\mathrm{jp}}\left(n\right)=\mathrm{\η}{\mathrm{\δ}}_p^P\left(n\right)v_j^J\left(n\right),$ ?w _jp(n+1)＝w _jp(n)+Δw _jp(n)j＝1，2，Λ，J；p＝1，2Λ，P (16)

$\mathrm{\Δ}{w}_{\mathrm{ij}}\left(n\right)=\mathrm{\η}{\mathrm{\δ}}_j^J\left(n\right)v_i^I\left(n\right),$ ?w _ij(n+1)＝w _ij(n)+Δw _ij(n)i＝1，2，Λ，I:j＝1，2，Λ，J (17)

$\mathrm{\Δ}{w}_{\mathrm{mi}}\left(n\right)=\mathrm{\η}{\mathrm{\δ}}_i^I\left(n\right)x_{\mathrm{km}}\left(n\right)$ ?w _mi(n+1)＝w _mi(n)+Δw _mi(n)m＝1，2，Λ，M；i＝1，2，Λ，I (18)

The 8th step judged whether to finish all training samples, was then to finish, otherwise went to for the 3rd step.

3, PCA﹠amp; BP (principal component analysis combines with the BP network) method modeling

3.1, sample set is defined as follows:

Input vector Ipb, output vector Opb, target vector are Tpb.

Wherein input vector Ipb is defined as the spectrum samples training data that collects.Target vector Tpb is defined as:

Tpb＝[1?2?3?4?5?6?7?8]；

Table 2:PCA﹠amp; The target substance of BP and the theoretical output valve table of comparisons

Target substance	Acetone	Diesel oil	Machine oil	Alcohol	Gasoline	Lubricating oil	Salad oil	Seawater
									Theoretical output valve	1	2	3	4	5	6	7	8

If real output value is 0.998 ≈ 1, judge that then this kind material is an acetone; If real output value is 6.027 ≈ 6, judge that then this kind material is a lubricating oil.

3.2, PCA﹠amp; The structure of BP method and principle

The oil spilling spectrum samples is at first passed through the principal component analysis algorithm as shown in Figure 1, the principal component analysis compression algorithm spectroscopic data of X dimension, each X dimension spectrum vector of compression back becomes the spectrum vector of 8 dimensions, this 8 dimension is exactly the principal component of X position spectrum vector, they are comprising the information of raw data more than 90%, data volume is compressed greatly simultaneously, and the individual character of outstanding each spectrum.Data after the compression are input to the BP network again carry out discriminator.Because data volume reduces greatly, the individual character of spectrum is distinct more simultaneously.So the recognition speed of BP network and precision are improved greatly, the time that whole process spent is more a lot of than only utilizing the BP Network Recognition to reduce.

It should be noted that because this method will utilize the BP neural network to discern at last, and the dimension of input vector is 8 dimensions.These data that just require to be identified also are the oil spilling spectroscopic datas of 8 dimensions.This just need be before identification, the spectrum vector that training spectrum vector sum is to be identified is formed a matrix, then this matrix is input in the PCA algorithm again, the principal component that not only extracts training spectrum vector also extracts the principal component that will be identified oil spilling sample spectrum vector simultaneously, so just can obtain the oil spilling sample vector to be identified of 8 dimensions.

Fig. 2 is a remote sensing monitoring oil spill area intelligent algorithm process flow diagram of the present invention, and wherein, the extraction of image feature value is as follows:

Choose the input of one 5 n dimensional vector n of gray-scale value composition of 4 eigenwerts (entropy, angle second moment, steadily local and non-similarity) of image texture and pixel as neural network.

During the textural characteristics value of computed image, select calculation window big more, the local message of estimation is accurate more.But big window makes split window lower in the precision of boundary different texture character homogenising.By further discovering, the original image of different resolution need adopt different window sizes, if image adopts 15 * 15 window size, selects different from d (d=1 in addition respectively, d=2, d=3) with different direction θ (0 °, 45 °, 90 °, 135 °) test, from test findings, when d=2, the effect when horizontal direction is 0 ° is best.

Determined to calculate after each parameter of textural characteristics value, just can utilize algorithm of co-matrix to obtain the image texture features value.At first, the representative region of choosing oil spilling, seawater, mountain region in image respectively is used for network training, and area size all is 90 * 90 pixels.Because the size of calculation window is 15 * 15, go out the texture eigenwert and compose so on the image of representative region, move this window calculation to the center pixel, form 5 dimensional feature vectors in conjunction with the gray-scale value of 4 characteristic quantities and center pixel then.The level of active window and vertical moving number of times all are 6 times, and each representative region is 36 groups of data, and totally 108 groups of data of 3 class materials have been formed one 108 * 5 input of tieing up matrix as neural metwork training.

Extract part in the eigenwert of emulating image and adopt identical method, just active window is that single pixel moves, the i.e. textural characteristics value of the intact pixel of every calculating in emulating image, to the right or move down a pixel, rather than move a window size resembling in training image.The purpose of doing like this is to guarantee in the emulating image that almost each pixel can both extract feature and be classified identification, the resolution of the back image that can guarantee to classify.And partly have several pixels can not be extracted out eigenwert in edge of image, also just can not be classified identification.

Because ultraviolet remote sensing can monitor thin oil film, and the infrared remote sensing image can monitor thicker oil film, therefore, the present invention can discern thicker oil film with infrared remote sensing image and ultraviolet image stack, and its classification than heavy oil film (area) is calculated identical with the intelligent algorithm of offshore spilled oil remote sensing monitoring area.

Portable airborne remote sensing of the present invention is monitored the marine oil overflow system and can be equipped in easily on the various ocean patrol aircrafts, does not need aircraft that power supply is provided.When the marine oil overflow accident takes place, this system's energy real-time follow-up marine oil overflow, estimation oil spillage and contaminated area are also removed effect to oil spilling and are assessed.Round-the-clock remote sensing Real time identification oil spilling, accuracy rate 〉=85%; All-weather real-time detects the oil spilling on sea ice, the snow, the oil spilling on the seabeach, accuracy rate 〉=85%; Observable oil film〉2 μ m; Observable minimum target is 5m radially.

Claims (1)

1. portable airborne offshore oil overflowing remote sense monitoring system, comprise infrared/ultraviolet/multispectral digital camera, central control and display system, data management and image processing operations software systems, described portable airborne offshore oil overflowing remote sense monitoring system is equipped on the ocean patrol aircraft, do not need aircraft that power supply is provided, need not aircraft is transformed; Described infrared/ultraviolet/multispectral digital camera is used for obtaining respectively the infrared/ultraviolet/multispectral digital image information on monitored sea; Control of described central authorities and display system are by portable industrial pc, GPS, GIS and system of image acquisition function software form, be used to realize by above-mentioned infrared/demonstration of the offshore oil overflowing remote sense surveillance map picture that ultraviolet/multispectral digital camera obtained, the storage of offshore oil overflowing remote sense surveillance map picture, the stack of offshore oil overflowing remote sense surveillance map picture and electronic chart, the playback of offshore oil overflowing remote sense surveillance map picture, the daily record of cruising, the management of oil spilling image file, the time recording of flight path, individual collection and the continuous acquisition of single-sensor, the collection in turn of multisensor, the record of video information, online flight path playback; Described data management and image processing operations software systems are installed in the portable industrial pc of central authorities control and display system, be used to read offshore oil overflowing remote sense monitoring image information and finish the ONLINE RECOGNITION oil spilling, at line computation contaminated area and the thick oil film of ONLINE RECOGNITION;

The identification of marine oil overflow is carried out in the method modeling that described portable airborne offshore oil overflowing remote sense monitoring system adopts the PCA principal component analysis to combine with the BP network, and its recognizer is as follows:

The oil spilling spectrum samples is at first passed through the principal component analysis algorithm, the principal component analysis compression algorithm spectroscopic data of X dimension, each X dimension spectrum vector of compression back becomes the spectrum vector of 8 dimensions, this 8 dimension is exactly the principal component of X position spectrum vector, they are comprising the information of raw data more than 90%, data volume is compressed greatly simultaneously, and the individual character of outstanding each spectrum, the data after the compression is input to the BP network again carries out discriminator; Before identification, the spectrum vector that training spectrum vector sum is to be identified is formed a matrix, then this matrix is input in the PCA algorithm again, the principal component that extracts training spectrum vector extracts the principal component that will be identified oil spilling sample spectrum vector simultaneously, obtains the oil spilling sample vector to be identified of 8 dimensions; Sample set is defined as follows: input vector Ipb, and output vector Opb, target vector are Tpb, wherein input vector Ipb is defined as the spectrum samples training data that collects, target vector Tpb is defined as Tpb=[1 234567 8]; Target substance is corresponding with theoretical output valve, i.e. acetone=1, and diesel oil=2, machine oil=3, alcohol=4, gasoline=5, lubricating oil=6, salad oil=7, the kind of target substance is judged in seawater=8 according to real output value and theoretical output valve;

The round-the-clock remote sensing Real time identification of described portable airborne offshore oil overflowing remote sense monitoring system oil spilling, accuracy rate 〉=85%; All-weather real-time detects the oil spilling on sea ice, the snow, the oil spilling on the seabeach, accuracy rate 〉=85%; Observable oil film＞2 μ m; Observable minimum target is 5m radially.

Images