CN103164853B - Imaging coupling assisting navigation can matching appraisal procedure - Google Patents

Abstract

What the present invention relates to a kind of imaging coupling assisting navigation can matching appraisal procedure, and step comprises: (1) is selected to be used for can the parameter set of matching assessment; (2) assessment result classification is determined; (3) use above-mentioned can matching assessment parameter set as input, assessment result as output, structural model identification artificial neural network; (4) priori training mode identification artificial neural network is used; (5) undertaken matching assessing by step (4) trained neural network, using the input of the above-mentioned parameter of reference map as pattern-recognition artificial neural network, export assessment result by pattern-recognition artificial neural network.Described parameter set comprises: reference map gray variance, correlation surface main peak value and main peak sharpness.The trouble that the method not only can avoid parameter threshold to select and parameter threshold select the improper risk easily causing erroneous judgement, and the priori that can make full use of constantly accumulation progressively improves the accuracy of assessment.

Description

Imaging coupling assisting navigation can matching appraisal procedure

Technical field

The invention belongs to aircraft assisting navigation technical field, particularly a kind of imaging coupling assisting navigation can matching appraisal procedure.

Technical background

Scene Matching Techniques utilizes imaging sensor (as imaging sensors such as visible ray, infrared, the radars) scene image (hereinafter referred to as real-time figure) of Real-time Obtaining presumptive area in aircraft flight process, with be pre-stored in carry-on reference image data (hereinafter referred to as reference map) and carry out registration, the absolute position current with calculating aircraft or the position data of relative target.Scene Matching Techniques has the advantages such as navigation accuracy is high, independence is strong, plays an important role in aircraft assisting navigation.In some applications, need preferred Scene matching area, in other application, then need to assess its reliability for scene matching aided navigation and precision for certain scene Matching band, in other words can matching, these can carry out based on characteristics such as the structure of reference map, textures.

At present, for visible ray, infrared imaging or radar imagery coupling assisting navigation aircraft, before the use, can assess it according to reference map can matching.For assessment of parameter comprise reference map gray variance, independent pixels, information entropy, correlation surface main peak value, primary and secondary peak ratio, main peak sharpness, the parameters such as feature scene quantity of information, complexity, excentricity.Carry out can matching assessment time, need the threshold range setting each parameter, parameter threshold select can only be obtained by statistics the impact of assessment result, not having dominant formula to represent, therefore be the reasonable threshold value scope of Selection parameter, can only pass through repeatedly sound out.Because parameter is many, this workload repeatedly soundd out is just very large.And once parameter threshold scope is determined, the evaluate knowledge through checking obtained afterwards can not automatically for improving the accuracy of assessment, except unartificial again exploration determines new parameter threshold scope, therefore, current imaging matching process threshold value is not easily determined, inefficiency.

Summary of the invention

The object of the invention is to, overcome the limitation of prior art, thus provide a kind of imaging based on Pattern Recognition mate assisting navigation can matching appraisal procedure.

For achieving the above object, a kind of imaging coupling assisting navigation provided by the invention can matching appraisal procedure, as shown in Figure 1, this can matching appraisal procedure step be:

(1) select be used for can matching assessment parameter set, alternative parameter comprises: reference map gray variance, independent pixels, information entropy, correlation surface main peak value, primary and secondary peak ratio, main peak sharpness, feature scene quantity of information, complexity and excentricity.The parameter such as general at least selection reference figure gray variance, correlation surface main peak value, main peak sharpness during selected parameter.For improving the accuracy of assessment, multiselect parameter of can trying one's best, but multiselect parameter can increase the time of calculating parameter and the input node of artificial neural network.

(2) assessment result classification is determined, assessment result can be can not mate maybe can mate (can represent with 0 and 1 respectively), also can be can not mate, can mate but error compared with maybe mating and error little (can represent with 0,1 and 2 respectively) greatly, can also be can not mate, can mate but mismatch possibility and greatly maybe can mate and mismatch possibility little (also can represent with 0,1 and 2 respectively) etc.

(3) use above-mentioned can matching evaluate parameter collection as input, assessment result as output, structural model identification artificial neural network.Network adopts BP(Back-Propogation, backpropagation) Multi-layered Feedforward Networks of learning algorithm, each input node of network corresponds to a parameter of sample, and output layer nodes equals assessment result classification number, a corresponding class of output node.

(4) priori neural network training is used.To fly or l-G simulation test obtains imaging and schemes in real time by hanging, and to mate with corresponding reference map, artificial interpretation confirmation is carried out to the accuracy of coupling and precision; Calculate each reference map above-mentioned parameter and as a sample; Corresponding matching result is also used for the training of this artificial neural network by the sample of accumulation some, generally 80% of sample number can be used for training, 20% and be used for verification, when neural network is by repeatedly training its assessment correctness still can not reach requirement, the number of hidden nodes of adjustment neural network re-starts training until its assessment correctness reaches requirement.

(5) trained neural network can carry out matching assessing.When assessing, the above-mentioned parameter of Calculation Basis figure, it can be used as the input of neural network, and the output of neural network is assessment result.

The invention has the advantages that, imaging of the present invention coupling assisting navigation can matching appraisal procedure owing to introducing Pattern Recognition, being used for without the need to artificial setting can the threshold value of parameter of matching assessment, and newly-increased matching knowledge can be used for further training mode identification artificial neural network, to improve the reliability that it differentiates further.

Accompanying drawing explanation

Fig. 1 be imaging of the present invention coupling assisting navigation can the schematic flow sheet of matching appraisal procedure;

Fig. 2 is the structural representation of the artificial neural network that the present invention adopts;

Fig. 3 a is that an embodiment of the inventive method carries out artificial neural network training result schematic diagram;

Fig. 3 b is the result schematic diagram that the embodiment of the inventive method carries out artificial neural network training;

Fig. 3 c is the assay schematic diagram that the embodiment of the inventive method carries out artificial neural network training;

Fig. 3 d is the overall result schematic diagram that the embodiment of the inventive method carries out artificial neural network training.

Embodiment

In order to more fully understand the present invention, and for understanding objects and advantages of the present invention, in conjunction with respective drawings and embodiment, the present invention is described in detail now.

Radar scene can matching assessment in, selection reference figure gray variance, independent pixels, information entropy, correlation surface main peak value, primary and secondary peak ratio, main peak sharpness, the parameters such as feature scene quantity of information, complexity, excentricity, as can matching evaluate parameter collection, namely input as neural network.Wherein,

(1) independent pixels is defined as follows:

Independent image prime number N is defined as:

$N=\frac{W}{L_x}\×\frac{H}{L_y}$

In formula:

W---picture traverse; H---picture altitude; L _xand L _ybe respectively the persistence length in x and y direction, persistence length is the displacement increment of coefficient of autocorrelation when dropping to 0.368.

(2) information entropy

If the gray level that all pixels of piece image have is set { b _i, and certain gray level b _ithe probability occurred is P (b _i), so average information of this width image, namely the computing formula of information entropy is:

$H=-\underset{i=1}{\overset{L}{\mathrm{\Σ}}}P\left(b_i\right)\·{\log }_{2}P\left(b_i\right);$

(3) correlation surface main peak sharpness

Correlation surface adopts normalized crosscorrelation algorithm to calculate.

In main peak sharpness correlation surface, the average of main peak surrounding 8 pixels characterizes with the ratio of main peak value.

(4) feature scene quantity of information

For bianry image, quantity of information can calculate with the ratio of the area of feature scene with the reference map total area, and the pixel number that simpler method is statistical nature scene calculates.

(5) feature scene complexity

For bianry image, the complexity of feature scene is reflected by the girth of feature scene and area usually.Suppose the girth of P representation feature scene, the area of A representation feature scene, so the computing formula of feature scene complexity C is: C=P ²/ A.

(6) feature scene excentricity

For bianry image, the computing formula of excentricity is:

$p/q=\frac{\sqrt{2/[(A+B)+\sqrt{{(A-B)}^2+4H^2}]}}{\sqrt{2/[(A+B)-\sqrt{{(A-B)}^2+4H^2}]}}---\left(17\right)$

In formula: $A=\mathrm{\Σ}{m}_i(y_i^2+z_i^2);$ $B=\mathrm{\Σ}{m}_i(x_i^2+z_i^2)$ The moment of inertia around X, Y-coordinate axle respectively; H=∑ m _ix _iy _ifor the product of inertia, m _ifor bianry image pixel value.

Assessment result, for mating and can not mate, represents with 1 and 2 respectively.

As shown in Figure 2, construct 9 inputs, 2 BP artificial neural networks exported, initial setting the number of hidden nodes is 10.

Obtain 57 samples, namely obtain the above-mentioned parameter of reference map corresponding to 57 Scene matching area, and demonstrate by emulation mode and hang the real-time diagram data flying to obtain and can mate at these Matching band.Neural metwork training is used at random from these 57 samplings 42 samples, training result for have 32 (accounting for 76.2% of training sample) can Matching band (representing with 1) be identified as can Matching band, 0 (accounting for 0.0% of training sample) can Matching band be identified as can not Matching band (representing with 2), 4 (accounting for 9.5% of training sample) can not Matching band be identified as can Matching band, 6 (accounting for 14.3% of training sample) can not Matching band be identified as can not Matching band, recognition correct rate is 90.5%, False Rate is 9.5%, as shown in Figure 3 a; Randomly draw 9 verifications for neural network from remaining 15, recognition correct rate is 88.9%, and False Rate is 11.1%, as shown in Figure 3 b; Remaining 6 samples are used for test, and recognition correct rate is 100%, and False Rate is 0%, as shown in Figure 3 c; Training and test the long and are shown in Fig. 3 d.As can be seen from Fig. 3 a ~ 3d, neural network has 4 samples to be mistaken for can not to mate for mating in the training stage, has 1 can mate be mistaken for and can not mate in the verification stage, and at test phase, then all interpretation is accurate.

Of the present invention above-mentioned for visible ray, infrared imaging or radar imagery coupling assisting navigation aircraft, before the use, can assess it according to reference map can matching.For assessment of parameter comprise reference map gray variance, independent pixels, information entropy, correlation surface main peak value, primary and secondary peak ratio, main peak sharpness, the parameters such as feature scene quantity of information, complexity, excentricity.Carry out can matching assessment time, because parameter is many, the threshold range of parameter selects difficulty.For this reason, propose based on Pattern Recognition can matching assessment technology, using for assessment of parameter set as the input of pattern-recognition artificial neural network, assessment result exports as it, by hang fly or Simulation results accumulation priori for the training of this artificial neural network and verification, through this artificial neural network that repetition training, School Affairs are optimized and revised, namely can be used for can matching assessment.The trouble that the method not only can avoid parameter threshold to select and parameter threshold select the improper risk easily causing erroneous judgement, and the priori that can make full use of constantly accumulation progressively improves the accuracy of assessment.

It should be noted last that, above embodiment is only in order to illustrate technical scheme of the present invention and unrestricted.Although with reference to embodiment to invention has been detailed description, those of ordinary skill in the art is to be understood that, modify to technical scheme of the present invention or equivalent replacement, do not depart from the spirit and scope of technical solution of the present invention, it all should be encompassed in the middle of right of the present invention.

Claims (4)

1. imaging coupling assisting navigation can a matching appraisal procedure, step comprises:

(1) select be used for can matching assessment parameter set; Described parameter set comprises: reference map gray variance, correlation surface main peak value and main peak sharpness;

(2) assessment result classification is determined;

(3) use above-mentioned can matching assessment parameter set as input, assessment result as output, structural model identification artificial neural network;

(4) priori training mode identification artificial neural network is used;

The process of described use priori neural network training is: to fly or l-G simulation test obtains imaging and schemes in real time by hanging, and to mate with corresponding reference map, carries out artificial interpretation confirmation to the accuracy of coupling and precision; Calculate each reference map above-mentioned parameter and as a sample; Corresponding matching result is also used for the training of this artificial neural network by the sample of accumulation some, be used for training, 20% by 80% of sample number and be used for verification, when neural network is by repeatedly training its assessment correctness still can not reach requirement, the number of hidden nodes of adjustment neural network re-starts training until its assessment correctness reaches requirement;

(5) undertaken matching assessing by step (4) trained neural network, using the input of the above-mentioned parameter of reference map as pattern-recognition artificial neural network, export assessment result by pattern-recognition artificial neural network.

2. imaging coupling assisting navigation as claimed in claim 1 can matching appraisal procedure, it is characterized in that, described parameter set also comprises further: independent pixels, information entropy, primary and secondary peak ratio, feature scene quantity of information, complexity and excentricity.

3. imaging coupling assisting navigation as claimed in claim 1 can matching appraisal procedure, it is characterized in that, described assessment result classification comprises: can not mate and maybe can mate; Can not mate, can mate but error compared with maybe can to mate greatly and error is little; Can not mate, can mate but mismatch possibility and greatly maybe can to mate and to mismatch possibility little.

4. imaging coupling assisting navigation as claimed in claim 1 can matching appraisal procedure, it is characterized in that, described pattern-recognition artificial neural network adopts the Multi-layered Feedforward Networks of back propagation learning algorithm, each input node of network corresponds to a parameter of sample, and output layer nodes equals assessment result classification number, a corresponding assessment result classification of output node.