CN104077535A - Graphic information system (GIS) vector data local decryption and restoring method - Google Patents

Abstract

The invention discloses a local decryption and recovery method of GIS vector data, which includes the following process: (1) calculate the decryption parameters through the local decryption control points, and iterate the parameters to control the medium error in the decryption area; (2) according to The local declassification parameters are used to declassify the local elements; (3) according to the local declassification parameters, the declassified local elements are restored. The method of the present invention has the characteristics of progressive partial declassification deformation, precise transformation of declassification control points and tight support of densification areas, and can meet the requirement of local declassification processing of GIS vector data for external release and use.

Description

The local DecryptDecryption of a kind of GIS vector data and restoration methods

Technical field

The invention belongs to geography information security fields, be specifically related to the local DecryptDecryption of a kind of GIS vector data and restoration methods.

Background technology

Restrictive condition and the secure contents of file to open map and geography information such as " regulation of Surveying Management Work state secret scope ", " open map content representation supplementary provisions (trying) " and " regulation (trying) of Fundamental Geographic Information System demonstration content " have been made regulation, and the confidentiality of the spatial position precision of lay special stress on Fundamental Geographic Information System and relevant factor.

Common facility and responsive facility have different positional accuracy index requests.To local sensitivity key element carry out local geometric precision DecryptDecryption need to be by its accurate DecryptDecryption to target location to meet positional accuracy requirement.In addition when responsive key element is carried out to local DecryptDecryption, should not exert an influence to the key element outside certain limit, i.e. the tight support performance of local DecryptDecryption.Existing local DecryptDecryption model adopts sectional pattern conventionally, will be on a large scale map partitioning among a small circle, re-use the model such as polynomial expression or rubber transformation of page and carry out Local treatment, be generally difficult to meet reference mark and accurately convert, convert evenly asymptotic and three demands of the tight support of deformed region.

Summary of the invention

The present invention is directed to the defect that the local DecryptDecryption model of existing GIS vector data exists, propose a kind of local DecryptDecryption method of GIS vector data based on Compactly supported radial basis function, have the advantages that distortion is progressive, reference mark accurately converts and deformed region tightly supports.

The technical solution used in the present invention is as follows:

The local DecryptDecryption of a kind of GIS vector data and restoration methods, comprise following process:

(1) key generative process

Step 11: determine DecryptDecryption scope

The minimum boundary rectangle RECTANGLE in input DecryptDecryption region, wherein, rectangle RECTANGLE lower left corner coordinate is (x _min, y _min), upper right corner coordinate is (x _max, y _max), obtain data directions X length X L and Y-direction length YL according to formula (1);

$\left\{\begin{array}{c}\mathrm{XL}=X_{\max }-X_{\min }\\ \mathrm{YL}=Y_{\max }-Y_{\min }\end{array}\right.---\left(1\right)$

Step 12: specified data reference mark and DecryptDecryption converted quantity

Input source reference mark set fromPoints={ (Fx _i, Fy _i) | i=1,2 ... k} and target control point set toPoints={ (Tx _i, Ty _i) | i=1,2 ..., k} forms k dominating pair of vertices, meets the condition that does not all contain coincide point in reference mark, source set fromPoints and target control point set toPoints;

Step 13: the radius of influence R of determining each DecryptDecryption reference mark according to formula (2);

$R\≥100\×\max \left(\sqrt{{({\mathrm{Tx}}_i-{\mathrm{Fx}}_i)}^2+{({\mathrm{Ty}}_i-{\mathrm{Fy}}_i)}^2}\right)---\left(2\right)$

Step 14: training Compactly supported radial basis function neural network model

A) selecting single order tightly to support basis function as output function Φ, is c (c for center _x, c _y) basis function, it is at data point p _iplace is output as:

Select the distance of Euclidean distance computational data point p and center c;

$\mathrm{dis}\tan \mathrm{ce}(p,c)=\left|\right|p-c\left|\right|=\sqrt{{(p_x-c_x)}^2+{(p_y-c_y)}^2}---\left(4\right)$

B) with reference mark, source set fromPoints coordinate (Fx _i, Fy _i) as input layer X, DecryptDecryption converted quantity (Tx _i-Fx _i, Ty _i-Fy _i) as output layer learning sample y, the individual hidden node of k is as the center c={ (Fx of basis function in hidden layer _i, Fy _i) | i=1,2 ... k}, each basis function is got identical R value, the output composition matrix H of k hidden node, and hidden layer to the neuron between output layer connects weights W, set up Compactly supported radial basis function neural network CSRBFnet, meet the interpolation condition of formula (5);

y＝HW (5)

C) utilize least square method to calculate weights W by formula (6), W is 2*k matrix;

W＝H ^-1y (6)

D) within the scope of minimum boundary rectangle RECTANGLE, evenly choose m*n sample point composition sample point S set amplePoints={ (Sx _j, Sy _j) | j=1,2 ..., num}, wherein m is directions X sample point quantity, n is Y-direction sample point quantity, m>=3, n>=3, num=m*n; Travel through sample point S set amplePoints, calculate the disturbance quantity of each sample point according to formula (7), generate sample point disturbance quantity S set amplePoints '={ (Sx _j', Sy _j') | j=1,2 ..., num};

$\left\{\begin{array}{c}{\mathrm{Sx}}_i={\mathrm{\Σ}}_{l=1}^k{w}_{1l}{\mathrm{\Φ}}_l\left(\mathrm{dis}\tan \mathrm{ce}(X_i,c_l)\right)\\ {\mathrm{Sy}}_i={\mathrm{\Σ}}_{l=1}^k{w}_{2l}{\mathrm{\Φ}}_l\left(\mathrm{dis}\tan \mathrm{ce}(X_i,c_l)\right)\end{array}\right.---\left(7\right)$

E) according to error RMSE in formula (8) calculating;

$\mathrm{RMSE}=\sqrt{\frac{\mathrm{\Σ}(s{x}_i^2+s{y}_i^2)}{\mathrm{num}}}---\left(8\right)$

If f) | offset/RMSE-1|>0.01, do not meet the requirement of input data global transformation amount offest, use the radius of influence R at the each DecryptDecryption of formula (9) iteration reference mark to carry out convergent-divergent with error of centralization RMSE, circulation step b-e, until | offset/RMSE-1|<=0.01, for having resolved;

$R=R\&times;\frac{\mathrm{offset}}{\mathrm{RMSE}}---\left(9\right)$

D) use radius of influence R and the weight matrix W at reference mark, source set fromPoints, target control point set toPoints, each DecryptDecryption reference mark to form DecryptDecryption parameter K ey, use RSA Algorithm to carry out asymmetric encryption and deposit key file Key.txt in key K ey;

(2) local DecryptDecryption process

Step 21: read DecryptDecryption parameter K ey, after using RSA Algorithm deciphering, extract key K ey, the radius of influence R and the weight matrix W that obtain reference mark, source set fromPoints, target control point set toPoints, each DecryptDecryption reference mark, set up DecryptDecryption Compactly supported radial basis function neural network CSRBFnet;

Step 22: open and treat DecryptDecryption vector data D, extract the key element point coordinate of vector data D, obtain key element point coordinate set P={ (x _j, y _j) | j=1,2 ..., k}, wherein k is the some number that key element comprises;

Step 23: by the Compactly supported radial basis function neural network CSRBFnet having set up, according to each point coordinate p in formula (10) computational element point coordinate set P _j(x _j, y _j) through changes in coordinates amount (the Δ x after DecryptDecryption conversion _j, Δ y _j);

$\left\{\begin{array}{c}\mathrm{\&Delta;}{x}_j={\mathrm{CSRBFnet}}_x\left(p_j\right)={\mathrm{\&Sigma;}}_{l=1}^k{w}_{1l}{\mathrm{\&Phi;}}_l\left(\mathrm{dis}\tan \mathrm{ce}(p_j,c_l)\right)\\ \mathrm{\&Delta;}{y}_j={\mathrm{CSRBFnet}}_y\left(p_j\right)={\mathrm{\&Sigma;}}_{l=1}^k{w}_{2l}{\mathrm{\&Phi;}}_l\left(\mathrm{dis}\tan \mathrm{ce}(p_j,c_l)\right)\end{array}\right.---\left(10\right)$

Step 24: DecryptDecryption changes in coordinates amount is applied to point coordinate p according to formula (11) _j, obtain point coordinate set P '={ (x _j', y _j') | j=1,2 ..., k};

$\left\{\begin{array}{c}{x}_j^{\&prime;}=x_j+\mathrm{\&Delta;}{x}_j\\ y_j^{\&prime;}=y_j+\mathrm{\&Delta;}{y}_j\end{array}\right.---\left(11\right)$

Step 25: circulation step 23 and 24, until all key elements are disposed, preserve the data file DF after local DecryptDecryption;

(3) local recovery's process

Step 31: read key file Key.txt, after using RSA Algorithm deciphering, extract key K ey, obtain reference mark, source and the set of target control point, radius of influence R and the weight matrix W at each local DecryptDecryption reference mark, set up DecryptDecryption Compactly supported radial basis function neural network CSRBFnet;

Step 32: open the data DF after DecryptDecryption, extract the key element point coordinate of vector data DF, obtain key element point coordinate set the P '={ (Px after DecryptDecryption _i', Py _i') | i=1,2 ..., k}.And the DecryptDecryption converted quantity of supposing each point is Δ _i={ (Δ x _i, Δ y _i) | i=1,2 ..., k}, initial value is 0;

Step 33: according to formula (12), by p '-Δ _p' substitution neural network CSRBFnet, upgrades the DecryptDecryption converted quantity Δ of each point _i(Δ x _i, Δ y _i);

$\left\{\begin{array}{c}\mathrm{\&Delta;}{x}_i=P{x}_i^{\&prime;}-{\mathrm{CSRBFnet}}_x(p_i^{\&prime;}-{\mathrm{\&Delta;}}_i)\\ \mathrm{\&Delta;}{y}_i=P{y}_i^{\&prime;}-{\mathrm{CSRBFnet}}_y(p_i^{\&prime;}-{\mathrm{\&Delta;}}_i)\end{array}\right.---\left(12\right)$

Step 34: specification error limit value ∈, if the Δ after upgrading _imeet formula (13), be considered as having recovered, otherwise repeating step 33;

$\left\{\begin{array}{c}P{x}_i^{\&prime;}-{\mathrm{CSRBFnet}}_x(p_i^{\&prime;}-{\mathrm{\&Delta;}}_i)-{\mathrm{\&Delta;x}}_i<\&Element;\\ y_i^{\&prime;}-{\mathrm{CSRBFnet}}_y(p_i^{\&prime;}-{\mathrm{\&Delta;}}_i)-{\mathrm{\&Delta;y}}_i<\&Element;\end{array}\right.---\left(13\right)$

Step 35: repeating step 33 and 34, process successively each key element, the data file RF after saving/restoring.

The present invention proposes a kind of method of carrying out local DecryptDecryption and recovery for GIS vector data.This method is being safeguarded under the prerequisite that vector data topological relation is constant, DecryptDecryption reference mark accurately can be transformed to target control point, ensures the tight support performance in DecryptDecryption region simultaneously, can meet the local DecryptDecryption of GIS vector data and process externally to issue user demand.

Brief description of the drawings

Fig. 1 is key product process figure in the technology of the present invention;

Fig. 2 is the local DecryptDecryption process flow diagram of data in the technology of the present invention;

Fig. 3 is Data Recovery Process figure after local DecryptDecryption in the technology of the present invention;

Fig. 4 is the original vector data that the embodiment of the present invention is selected;

Fig. 5 is the design sketch of data stack after raw data and local DecryptDecryption in the embodiment of the present invention;

Fig. 6 is the local amplification effect figure of data stack after raw data and local DecryptDecryption in the embodiment of the present invention;

Fig. 7 is the design sketch that recovers data stack in the embodiment of the present invention after raw data and local DecryptDecryption.

Embodiment

Below in conjunction with drawings and Examples, the present invention is described in further details.

The present embodiment is selected shapefile form vector data, to data read, DecryptDecryption and recovery operation, further describe the present invention.The present embodiment selects the shapefile face figure layer data (as Fig. 4) in a certain area as original vector data.

(1) key generative process

Step 11: determine DecryptDecryption scope, the minimum boundary rectangle RECTANGLE in input DecryptDecryption region, RECTANGLE lower left corner coordinate is (164417.097000,162667.721400), upper right corner coordinate is (171415.635650,167846.074000), obtain data directions X length X L=6998.53865m, Y-direction length YL=5178.3526m;

Step 12: input reference mark, 6 pairs of sources and the set of target control point;

Step 13: determine the radius of influence R at each local DecryptDecryption reference mark, select R=3000m;

Step 14: training radial basis function neural network, concrete steps are as follows:

A) select single order tightly to support basis function as output function Φ;

B) with fromPoints coordinate (Fx _i, Fy _i) as input layer X, DecryptDecryption converted quantity (Tx _i-Fx _i, Ty _i-Fy _i) as output layer learning sample y, 6 hidden nodes are as the center c={ (Fx of basis function in hidden layer _i, Fy _i) | i=1,2 ... 6}, each basis function is got identical radius of influence R, and each basis function is output as H, and hidden layer to the neuron between output layer connects weights W, sets up radial basis function neural network CSRBFnet;

C) calculate weights W=[-0.68576558 2.24318105 1.30650649-0.924589651.0799589 1.09748868 by formula (6); 3.21258749 0.86481197-0.68393027 3.34310239-1.41045726-1.1620549];

D) within the scope of minimum boundary rectangle RECTANGLE, evenly choose 50*50 sample point composition sample point S set amplePoints={ (Sx _j, Sy _j) | j=1,2 ..., 2500}, travels through sample point S set amplePoints, calculates the disturbance quantity of each sample point according to formula (4), generates sample point disturbance quantity set SamplePoints '={ (Sx _j', Sy _j') | j=1,2 ..., 2500};

E) according to error RMSE=0.849401552583 in formula (8) calculating;

F), due to | offset/RMSE-1|>0.01, need radius of influence R=4158.10214553 at the local DecryptDecryption of iteration reference mark.Then circulation step b-e, R=3706.05059411 after iteration 2 times, RMSE=0.993166516737, satisfies condition, and obtains final weight matrix W=[-1.84601557 2.76989104 0.95848037-2.20952748 1.79216251 1.80692918; 4.42896143-0.15784362-1.20732732 4.89798742-2.70956928-2.02686662];

G) use radius of influence R and the weight matrix W at reference mark, source, the set of target control point, local DecryptDecryption reference mark to form local DecryptDecryption parameter K ey;

(2) local DecryptDecryption process

Step 21: read DecryptDecryption parameter K ey, after using RSA Algorithm deciphering, extract key K ey, the radius of influence R and the weight matrix W that obtain reference mark, source set fromPoints, target control point set toPoints, each DecryptDecryption reference mark, set up DecryptDecryption Compactly supported radial basis function neural network CSRBFnet;

Step 22: open original vector data D, extract the key element point coordinate of vector data D, obtain key element point coordinate set P={ (x _j, y _j) | j=1,2 ..., k}, wherein k is the some number that key element comprises, and describes for example below with 1 p (167131.647,164958.147) in P;

Step 23: set up Compactly supported radial basis function neural network CSRBFnet according to key K ey, calculate each point coordinate p in P set by formula (10) _j(x _j, y _j) variable quantity (Δ x _j, Δ y _j), the variable quantity of some p is (1.19198160762,2.25882888931);

Step 24: according to formula (11), DecryptDecryption changes in coordinates amount is applied to P, obtains point coordinate set P ', the DecryptDecryption recoil of some p is designated as (167132.838982,164960.405829);

Step 25: circulation step 23 and 24, until all key elements are disposed, preserve the data file DF after local DecryptDecryption;

(3) local recovery's process

Step 31: read key file Key.txt, after using RSA Algorithm deciphering, extract key K ey, obtain reference mark, source and the set of target control point, the radius of influence R at each DecryptDecryption reference mark and weight matrix W, set up local DecryptDecryption Compactly supported radial basis function neural network CSRBFnet;

Step 32: open the data DF after DecryptDecryption, extract the key element point coordinate of vector data DF, obtain key element point coordinate set the P '={ (Px after DecryptDecryption _i', Py _i') | i=1,2 ..., k}.And the DecryptDecryption converted quantity of supposing each point is Δ _i={ (Δ x _i, Δ y _i) | i=1,2 ..., k}, initial value is 0, below with 1 p ' (167132.838982,164960.405829) in P ' for example describes, Δ _p'=(0,0);

Step 33: according to formula 12, by p '-Δ _p' substitution RBFnet, the DecryptDecryption converted quantity Δ of renewal p ' _p'=(1.19416476507,2.25919063147);

Step 34: get and recover precision ∈=0.01m, the Δ after renewal _ido not meet 13, the 2 Δs of formula _p'=(1.1919782225,2.25883077318), meet formula 13, be considered as having recovered;

Step 35: repeating step 33 and 34, process successively each key element, the data file RF after saving/restoring.

In the embodiment of the present invention, only carry out DecryptDecryption and recovery operation as an example of shp formatted data example, the method also can be used for the extended formatting vector datas such as Geodatabase.

The foregoing is only the preferred embodiments of the present invention, be not limited to the present invention.Obviously, those skilled in the art can carry out various changes and modification and not depart from the spirit and scope of the present invention the present invention.Like this, if these amendments of the present invention and within modification belongs to the scope of the claims in the present invention and equivalent technologies thereof, the present invention is also intended to comprise these changes and modification interior.

Claims (1)

1. the local DecryptDecryption of GIS vector data and a restoration methods, comprises following process:

(1) key generative process

Step 11: determine DecryptDecryption scope

The minimum boundary rectangle RECTANGLE in input DecryptDecryption region, wherein, rectangle RECTANGLE lower left corner coordinate is (x _min, y _min), upper right corner coordinate is (x _nax, y _max), obtain data directions X length X L and Y-direction length YL according to formula (1);

$\left\{\begin{array}{c}\mathrm{XL}=X_{\max }-X_{\min }\\ \mathrm{YL}=Y_{\max }-Y_{\min }\end{array}\right.---\left(1\right)$

Step 12: specified data reference mark and DecryptDecryption converted quantity

Input source reference mark set fromPoints={ (Fx _i, Fy _i) | i=1,2 ... k} and target control point set toPoints={ (Tx _i, Ty _i) | i=1,2 ..., k} forms k dominating pair of vertices, meets the condition that does not all contain coincide point in reference mark, source set fromPoints and target control point set toPoints;

Step 13: the radius of influence R of determining each DecryptDecryption reference mark according to formula (2);

$R\&GreaterEqual;100\&times;\max \left(\sqrt{{({\mathrm{Tx}}_i-{\mathrm{Fx}}_i)}^2+{({\mathrm{Ty}}_i-{\mathrm{Fy}}_i)}^2}\right)---\left(2\right)$

Step 14: training Compactly supported radial basis function neural network model

A) selecting single order tightly to support basis function as output function Φ, is c (c for center _x, c _y) basis function, it is at data point p _iplace is output as:

Select the distance of Euclidean distance computational data point p and center c;

$\mathrm{dis}\tan \mathrm{ce}(p,c)=\left|\right|p-c\left|\right|=\sqrt{{(p_x-c_x)}^2+{(p_y-c_y)}^2}---\left(4\right)$

B) with reference mark, source set fromPoints coordinate (Fx _i, Fy _i) as input layer X, DecryptDecryption converted quantity (Tx _i-Fx _i, Ty _i-Fy _i) as output layer learning sample y, the individual hidden node of k is as the center c={ (Fx of basis function in hidden layer _i, Fy _i) | i=1,2 ... k}, each basis function is got identical R value, the output composition matrix H of k hidden node, and hidden layer to the neuron between output layer connects weights W, set up Compactly supported radial basis function neural network CSRBFnet, meet the interpolation condition of formula (5);

y＝HW (5)

C) utilize least square method to calculate weights W by formula (6), W is 2*k matrix;

W＝H ^-1y (6)

D) within the scope of minimum boundary rectangle RECTANGLE, evenly choose m*n sample point composition sample point S set amplePoints={ (Sx _j, Sy _j) | j=1,2 ..., num}, wherein m is directions X sample point quantity, n is Y-direction sample point quantity, m>=3, n>=3, num=m*n; Travel through sample point S set amplePoints, calculate the disturbance quantity of each sample point according to formula (7), generate sample point disturbance quantity S set amplePoints '={ (Sx _j', Sy _j') | j=1,2 ..., num};

$\left\{\begin{array}{c}{\mathrm{Sx}}_i={\mathrm{\&Sigma;}}_{l=1}^k{w}_{1l}{\mathrm{\&Phi;}}_l\left(\mathrm{dis}\tan \mathrm{ce}(X_i,c_l)\right)\\ {\mathrm{Sy}}_i={\mathrm{\&Sigma;}}_{l=1}^k{w}_{2l}{\mathrm{\&Phi;}}_l\left(\mathrm{dis}\tan \mathrm{ce}(X_i,c_l)\right)\end{array}\right.---\left(7\right)$

E) according to error RMSE in formula (8) calculating;

$\mathrm{RMSE}=\sqrt{\frac{\mathrm{\&Sigma;}(s{x}_i^2+s{y}_i^2)}{\mathrm{num}}}---\left(8\right)$

If f) | offset/RMSE-1|>0.01, do not meet the requirement of input data global transformation amount offest, use the radius of influence R at the each DecryptDecryption of formula (9) iteration reference mark to carry out convergent-divergent with error of centralization RMSE, circulation step b-e, until | offset/RNSE-1|<=0.01, for having resolved;

$R=R\&times;\frac{\mathrm{offset}}{\mathrm{RMSE}}---\left(9\right)$

D) use radius of influence R and the weight matrix W at reference mark, source set fromPoints, target control point set toPoints, each DecryptDecryption reference mark to form DecryptDecryption parameter K ey, use RSA Algorithm to carry out asymmetric encryption and deposit key file Key.txt in key K ey;

(2) local DecryptDecryption process

Step 21: read DecryptDecryption parameter K ey, after using RSA Algorithm deciphering, extract key K ey, the radius of influence R and the weight matrix W that obtain reference mark, source set fromPoints, target control point set toPoints, each DecryptDecryption reference mark, set up DecryptDecryption Compactly supported radial basis function neural network CSRBFnet;

Step 22: open and treat DecryptDecryption vector data D, extract the key element point coordinate of vector data D, obtain key element point coordinate set P={ (x _i, y _i) | j=1,2 ..., k}, wherein k is the some number that key element comprises;

Step 23: by the Compactly supported radial basis function neural network CSRBFnet having set up, according to each point coordinate p in formula (10) computational element point coordinate set P _j(x _j, y _j) through changes in coordinates amount (the Δ x after DecryptDecryption conversion _j, Δ y _j);

$\left\{\begin{array}{c}\mathrm{\&Delta;}{x}_j={\mathrm{CSRBFnet}}_x\left(p_j\right)={\mathrm{\&Sigma;}}_{l=1}^k{w}_{1l}{\mathrm{\&Phi;}}_l\left(\mathrm{dis}\tan \mathrm{ce}(p_j,c_l)\right)\\ \mathrm{\&Delta;}{y}_j={\mathrm{CSRBFnet}}_y\left(p_j\right)={\mathrm{\&Sigma;}}_{l=1}^k{w}_{2l}{\mathrm{\&Phi;}}_l\left(\mathrm{dis}\tan \mathrm{ce}(p_j,c_l)\right)\end{array}\right.---\left(10\right)$

Step 24: DecryptDecryption changes in coordinates amount is applied to point coordinate p according to formula (11) _j, obtain point coordinate set P '={ (x _j', y _j') | j=1,2 ..., k};

$\left\{\begin{array}{c}{x}_j^{\&prime;}=x_j+\mathrm{\&Delta;}{x}_j\\ y_j^{\&prime;}=y_j+\mathrm{\&Delta;}{y}_j\end{array}\right.---\left(11\right)$

Step 25: circulation step 23 and 24, until all key elements are disposed, preserve the data file DF after local DecryptDecryption;

(3) local recovery's process

Step 31: read key file Key.txt, after using RSA Algorithm deciphering, extract key K ey, obtain reference mark, source and the set of target control point, radius of influence R and the weight matrix W at each local DecryptDecryption reference mark, set up DecryptDecryption Compactly supported radial basis function neural network CSRBFnet;

Step 32: open the data DF after DecryptDecryption, extract the key element point coordinate of vector data DF, obtain key element point coordinate set the P '={ (x after DecryptDecryption _j', y _j') | j=1,2 ..., k}; And the DecryptDecryption converted quantity of supposing each point is Δ _i={ (Δ x _i, Δ y _i) | i=1,2 ..., k}, initial value is 0;

Step 33: according to formula (12), by p '-Δ _p' substitution neural network CSRBFnet, upgrades the DecryptDecryption converted quantity Δ of each point _i(Δ x _i, Δ y _i);

$\left\{\begin{array}{c}\mathrm{\&Delta;}{x}_i=P{x}_i^{\&prime;}-{\mathrm{CSRBFnet}}_x(p_i^{\&prime;}-{\mathrm{\&Delta;}}_i)\\ \mathrm{\&Delta;}{y}_i=P{y}_i^{\&prime;}-{\mathrm{CSRBFnet}}_y(p_i^{\&prime;}-{\mathrm{\&Delta;}}_i)\end{array}\right.---\left(12\right)$

Step 34: specification error limit value ∈, if the Δ after upgrading _imeet formula (13), be considered as having recovered, otherwise repeating step 33;

$\left\{\begin{array}{c}P{x}_i^{\&prime;}-{\mathrm{CSRBFnet}}_x(p_i^{\&prime;}-{\mathrm{\&Delta;}}_i)-{\mathrm{\&Delta;x}}_i<\&Element;\\ y_i^{\&prime;}-{\mathrm{CSRBFnet}}_y(p_i^{\&prime;}-{\mathrm{\&Delta;}}_i)-{\mathrm{\&Delta;y}}_i<\&Element;\end{array}\right.---\left(13\right)$

Step 35: repeating step 33 and 34, process successively each key element, the data file RF after saving/restoring.