TWI226012B - Neural network correcting method for touch panel - Google Patents

Abstract

There is provided a neural network correcting method for touch panel. The touch panel is divided into plural blocks, and each block is configured with a correction point. The selected coordinate value obtained by measuring each selected correction point is used as an input signal, and the original coordinate value of each configured correction point is used as an output signal. Based on the relation between the input signal and the output signal, neural learning algorithm is employed to generate the block weighted parameter and pressure shift parameter for setting up update formula , furthermore use neural operation algorithm to determine the corrected coordinate value after calibration, thereby increasing the accuracy in selecting the touch signal.

Description

1226012 发明 Description of the invention: [Technical field to which the invention belongs] A network correction method, especially a method for establishing a correction type by re-recording the amount of pressure recorded by a nerve-like practice such as a touch panel or the like of the present invention, Initial _ Correction method that uses the network to perform different methods to improve the accuracy of the touch signal. [Previous technology] According to 'there are many products on the market with resistive _ panels, these touch panels themselves have a heterogeneous taste and non-touch structure, and sugar panels touch two different components of the panel material, both_ The coordinates are incomplete. Furthermore, the characteristics of products made in the process are not the same. The thief needs to make corrections to improve the accuracy of the weaving. At this stage, the calibration method used in the touch panel is mostly derived from a few selected points, and then the calibration of the _sing_ points is performed, such as National Patent Bulletin Nos. 2 8 2 1 2 3 and 2 9 5 6 4 7 and so on. This -computing architecture requires _fine_trigonometric operations. When miscellaneous M is selected, the calculation age of the entire ship is quite complicated, and there is a problem of slow calculation processing speed. If there are few miscellaneous selections, and there are large errors, it is necessary to improve it ^. SUMMARY OF THE INVENTION The main purpose of this month is to solve the above problems and provide a neural network correction method such as a touch panel that can improve the accuracy of a touch signal. In order to achieve the foregoing object, the present invention divides the touch panel into several blocks, and sets at least one correction point in each block, and measures the point selection coordinate value obtained by selecting each correction point as an input signal. The original coordinate values of the calibration points (displayed on the display) are used as the rotation signal 1226012, and the input signal and rotation output are subtracted by a neural network-like county. The relationship training learns the weight parameters and bias records of each block, and uses a neural network-like algorithm as a correction to find the corrected coordinate values to improve the accuracy of the point and control signals. The detailed description of the other quilting points of the present invention, and the detailed description of the actual rewards and the accompanying drawings, have been thoroughly understood. Of course, the present invention allows some differences in the arrangement of other parts, or the arrangement of other parts, but the selected embodiment is described in detail in the present specification's towel, and its structure is shown in the towel. [Embodiment] The touch panel of the present invention has _ and _ correction green, the touch surface is divided into several blocks, and at least one correction point is set in each district, and each correction point is obtained Select the coordinate value as the turn-in signal, and set the original coordinate value of each calibration point as the output signal H. If the way 7 is the relationship between the input signal and the output signal from the method, and according to the input signal and output signal The relationship training learns to obtain the weight parameter and partial age number of each block, and then uses a neural network-like algorithm to calculate the correction formula to obtain the corrected coordinate value. Here is an example to explain the following: The example-using the Matlab software simulation, the touch panel: divided into five by five matrix of a total of 25 blocks, and each block is set-correction point h As shown in the figure above. Mingling reads the second figure, which is a flow chart of the method for applying a neural network (NeuraI Ne_k) correction method such as the present invention. When the touch panel is in use, the first step is to determine whether to perform the correction. Twenty-five miscellaneous people are turning the warp materials, and they have recorded the pressure parameters (b, and partial 1226012 pressure parameters (b), and get their corrections to update the weight parameters (Wx, Wy) and bias parameters (b) of neural-like algorithms. , And find X Π = Wx * X + b Y 11 = Wy * Y + b — Yu Qing_Financial correction, the input signal of the _secret chain value is replaced by the official knowledge, and the _, _ roadshow algorithm is used to calculate the correct result. Coordinate values., T 'point knows ^ Twenty-five correction points Tη, and measured the selected coordinates ^ as the input position, and set each correction point Tn _ seat _ money New York number set, in ^ implementation of financial, The coordinate values of the calibration points set by each calibration point τn miscellaneous occupational health; please refer to Figure 3 ', the coordinates of each point selected, the original coordinates of each calibration point, and the coordinates after calibration are shown. η (without a random error amount of ± 〇〇5) is as follows: P1 = [0,0]; P2 = [1,0.1]; P3 = [2,0.2]; PΦ = [3,0.1]; P5 = [ 4,0] P ^ = [〇.1,1]; P7 = [1, 1]; P8 = [2,1]; P9 = [3,1]; P10 = [3.9,1]; P11 = [〇.2,2]; P12 = [1,2]; P13 = [2,2 ]; P1Φ = [3,2]; P15 = [3.8,2] P16 = [0.1,3]; P17 = [1,3]; P18 = [23]; P19 = [3? 3]; P20 = [ 3.9? 3] Κ1 = [0,4]; Κ2 = [1,3 · 9]; Κ3 = [2,3 · 8]; Κ and [3,3 · 9]; Ρ25 = [4,4] each The original coordinates Tn of the calibration point are as follows: ΎΗ〇, 〇ΙΎ2 = [1 ^ Τ6 = [0,1]; Τ7 = [1,1]; Τ8 = [2,1]; Τ9 = [3,1]; Τ10 = [4,1] Τ11 = [0,2]; Τ12 = [1,2]; Τ13 = [2,2]; Τ1Φ = [3,2]; Τ15 = [4,2] Τ16 = [0, 3]; T17 = [13]; T18 = [2,3]; T19 = [3,3]; Ί2 (Κ4,3] Ί21 = [0,4]; Τ22 = [1,4]; Τ23 = [ 2,4]; Τ24 ^ [3,4]; Τ25 = [4,4] Then use a neural network-like algorithm to calculate the relationship between the input signal and the output signal. The yield check is as follows: for i = l: l: 25 1226012 // 25-point measurement coordinates, with a random error amount of + >0.05; end W = [00]; b = [0];

[FW FbHeam ^ a (P; T (aXW, ba A〇lX // Use Matlab's existing leak_a〇 fonctico to find the weight of X // and the bias value (Fb)

Xn = FW * Pf + Fb; // Calculate the corrected χ-coordinate W = [00]; b = [〇]; [FW Fb] = leam—a (P, T (:, 2) ', W, b , L, 0.011); // Use Matlab's existing leak-a〇flinction to find the weight of Y (pw) // and the bias value (Fb)

Yn = FW * P ， + Fb; // γ coordinate after counting

El ^ rnn ((P (U) fT (: a)). A2 > ^^ 〃 Calculate the error between the selected coordinate value p and the correction point T E2 = sim (Garding (:, 1)). 8 2) ^ 咖啡 ((¥ 1 ^ (:, 2)). 8 2) // The error amount of the coordinate value (Xn ^ Yn) after the calculation and the correction point T is similar to that of the neural network (Neural Network) based on the input signal and output signal. The relationship training learns the weight parameters (Wx, Wy) = [0.0009 1.0373] and the bias parameter (b) = -0.0598 for each block, and the correction formula is:

Xn = Wx (0.0009) * X + b (-0.0598), Y n = Wy (1.0373) * Y + b (-0.0598) Multiply the selection coordinates of fifteen points by the weight parameter (Wx, Wy) = [0.0009 1.0373], and then add the bias parameter ⑻ = _α〇598 to get the new 25-point corrected coordinates (χη, γη). The data is as follows: (^ 11 ， 丫 111 > ^ [0.0935 ，》 0.1039]; _, ^ 112 > = [; 0.9946,0.0564;]; 1226012 (χη3, Υη3M1.9818,0.1149]; (χη4, Υη4) = [3.0615, α〇585]; (χη5,5η5Μ4 · 〇574Γ0 · 0824]; (Xη6, Υη6) = [〇 · 〇535, 0 · 9811]; (Xη7, Υη7Η 0.9439, 1.0155]; (Xη8, Υη8) = [2 · 492, 09421]; (Χη9, Υη9Μ3.0181, 0.995]; ς χη10, Υη10) = [4.00110, 1.015]; (Χη11, Υη11) = [0 · 13 $, 2.0011]; (χη12, Υη12) = [〇9951, 2.0346]; (Xη13, Υη13Η1.9706α.9772]; (ΧηΗΥη1 ^^ (χη15, Υη15Η3.85172.〇 (Χ) 2] ;; (Χη16 ^ (^ 17, Υη17Μ〇 · 9357, 3.015); (Xη18 strokes (Xη19? Υη19Η3Ό1723Ό324); (Xη209Υη20 (Xη219Υη21Η ^ .〇79〇Λ〇743); (^ 2, Υη22Η.994〇Α〇111); (^ (Χη24, Υη24) = [3.0506Α0 028]; (Χη25, Υη25) = [4.0459,4.0695] Add the square of the subtraction of the coordinate τη (including ± 0 · 05 random error) at each point to the square of the original coordinate τη of the correction point. It is set to 0.2934, and the corrected coordinates (χη, γη) of each point are added to the square of the original coordinate Tη subtracted from the corrected point, and the resulting error amount is 0.1951. It can be seen that the corrected coordinates (χη, Υη) are increased by 33.5 % Accuracy. The calibrated touch panel does not need to be calibrated in the normal operation mode. When the user clicks, the coordinate values measured will be substituted into the correction formula, and a neural network-like calculation will be performed. The algorithm uses a correction formula to obtain the corrected coordinates (χη, γη) to improve the accuracy of the touch signal. Please refer to Figure 4® and Figure 5, which are the second embodiment of the present invention. The touch panel is divided into five health blocks of A, B, C, D, and E according to the coordinates of the coordinates, and each block is provided with five young positive points and red fifteen health wires according to the X axis of the coordinates. Divided into blocks to perform the quasi-god-f turn to obtain the five-weighted surface (w) and parameters, and the weight parameter (W) and bias parameter (b) of the Jing neural algorithm. _ General operation type is not corrected, the measured X and Y coordinates are first at A, B 'C, D, E _, and then the weight table and bias parameter of the ship ’s block are established. In the correction formula, the new Xn and Yn values are obtained by operation and update using a neural-like algorithm. This implementation calculates the twenty-fifth of the Matlab __ panel, and corrects the 1226012 timing. First, click on the twenty-five corrections, point Tn, and measure the coordinates of each point ^ as the turn-in signal. Use the original coordinate values of the calibration points Tn as the starting position. Please refer to Figure 6, Figure 4 for each _ coordinate, each calibration county coordinate, and each corrected coordinate. The coordinates of each click are as follows ...

Al = [〇 〇]; Α2 = [1 〇 · 1]; Α3 == [2 〇 2]; A4 = p 〇 1]; A5 = _; The 5 coordinate points in the B block are ...

Bl- [〇.l 1]; B2- [1 1]; B3 = [2 1]; B4 = [3 1]; B5 = [3.9 1]; The five selected coordinate points in block C are:

Cl- [0.2 2]; C2- [1 2]; C3 = [2 2]; C4 = [3 2]; C5 = [3.8 2]; The five selected coordinate points in the D block are:

Dl = [〇.l 3]; D2 = [l 3]; D3 = [2 3]; D4 = [3 3]; D5 = [3.9 3]; The 5 coordinate points in the E block are · · E1 = [0 4]; E2 = [1 3.9]; E3 = [2 3.8]; E4 = [3 3.9]; E5 = [4 4] The original coordinates Tn of the respective fathers are as follows: The coordinates of the five correction points are: ΤΑ1 = [0 0]; ΤΑ2 = [1 0]; ΤΑ3 = [2 0]; ΤΑ4 = [3 0]; ΤΑ5 = [4 0]; 5 corrections in the Β block The coordinates of the points are: TB1 = [0 1]; TB2 = [1 1]; TB3 = [2 1]; TB4 = [3 1]; TB5 = [4 1]; The coordinates of the five correction points in the C block are: : TC1 = [0 2]; TC2 = [1 2]; TC3 = [2 2]; TC4 = [3 2]; TC5 = [4 2]; The coordinates of the five calibration points in block D are: TD1 = [0 3]; TD2 = [1 3]; TD3 = [2 3]; TD4 = [3 3]; TD5 = [4 3]; The coordinates of the 5 correction points in the E block are: ΤΕ1 = [0 4 ]; ΤΕ2 = [1 4]; ΤΕ3 = [2 4]; Τ4 = [3 4]; Τ5 = [4 4] Then use a neural network-like algorithm to calculate the relationship between the input signal and the output signal, and its program is as follows: 1226012 Α (ι, :) ρΑ (ι,: > ^ (4) ^ * ΐΗπά (1 >20; // 5-point measurement coordinates in the Α region, with a random error amount of + Ό.05 end W = [00 ]; [AWXAbX] = leam_a (A; TA (: 4); W, b4,0.〇l); // Use Matlab's existing leak_a〇fonction to find the X white of area A Weight // (AWX) and the bias value (AbX)

XnA = AWX * A, + AbX; // Calculate the X coordinate of the Λ region after correction W: [00];

[AWAbYJ = leam_a (A; TA (: 2); W ^ 1A〇ll); // Use Matlab's existing lean__a〇fi] nctiai to find the weight of area A // (AWY) and the bias value (AbY )

YnA two AWY * A, + AbY; // Calculate the Y coordinate of the correction and area tbri = l: l: 5 B (i,: > = BCi; > ^ (4) ^ * rdnd (l >20; / / B area of 5 points measurement coordinates, with a random error of + Ό.05 end W = [00];

b = [0]; [BWXBbX] ^ eam_a (B ,, TB (:, l); W, b, 1A〇l); // use] ^ 1 ^ existing 103111_ & 〇 & 111〇11〇11 To find the weight of the ugly area // (BWX) and the bias value (BbX)

XnB = BWX * B '+ BbX; // Calculate the coordinates of B in the corrected area, W = [00]; b = [0]; DBWYBbYHeam-aO'TOOTWAOll); // Use Matlab's existing leak_a〇fonction, find The weight of area B // (BWY) and bias value (BbY) 11 1226012

YnB = BWY * B, + BbY; // Calculate the Y coordinate of the B area after correction tbri = l: l: 5 C (i,: K: Ci;) f (4 ^ Taiid (iy20; // 5 of the C area Point measurement coordinates, with a random error amount of + -0.05 end W = [00]; [CWX〇3X] ^ eam_a (C; TC (: a) ^ // Use Matlab's existing leakLaOfimctiQii to find the area X of C Weight // (CWX) and bias value (CbX)

XnOCWX * C + CbX; // Calculate the X coordinate of the C area after correction · W = [〇〇]; ㈣]; [CWCbYHeam_a (C ,, TC (: 2) ', W, b, l, 0.011); / / Use Matlab's existing leak_a〇fonction to find the weight of the C region Y // (CWY) and the bias value (CbY)

YnG = CWY * C, + CbY; // Calculate the Y coordinate of the C area after correction fori = l: l: 5

// 5-point measurement coordinates in the D area, with a random error amount of + Ό · 〇5 end W 2 [00]; b = [〇]; pm 卿 = 1 · —a (D, TO :, l), , W, b, l, 0.01); // Use the existing Leam_afonction of Matlab to find the weight of D area X // (DWX) and bias value (DbX)

XnD = DWX * D ， + DbX; // Calculate the X coordinate of the D area after correction W = [00j; b = [0]; DDWYDbYHeam—aPVID ^ WWAOll); 12 1226012 // Use Matlab's existing leak_a〇ftuxtion, find Out of the D area X weight // (DWY) and bias value (DbY)

YnMDWYWDbY; // Calculate the Y coordinate of D area after correction Γ〇〇 = 1: 1: 5 E (i;) = ECi,:) ^ (4 ^ Tand (iy20; // 5-point measurement coordinates of E area, A random error amount with · κ〇 · 05 end W = [00]; DEWXEbX] = leam__a (P, TE (:, l) ,, W, b, l, 0.01);

// Use Matlab's existing leak_a〇fonction to find the weight of the E region X // ^ WX) and the bias value (EbX)

XnE = EWX * E '+ EbX; // Calculate the X coordinate of the corrected E area W = [00]; [EWYEbYHeamUEi ^ WWAOll); // Use Matlab's existing leak_a〇fiinction to find the weight of the Y area E // (EWY) and bias value (EbY)

YnE = EWY * E + EbY; // Calculate Y coordinate of E area after correction

ElA = sum ((A (: ^ TA (: ^ // Calculate the amount of error between the coordinate values A1 ~ 5 in area A and the correction point TA) EZA ^ sumpiiA-TACJD / ^ sumPnA-TAe)) / ^) // Calculate The amount of error between the ship's fine m (Xn ^ YnA) in the Λ area and the correction point TA /// Calculate the error amount of the B-point selection coordinates B1 ~ 5 and the correction point TB E2B = sum ((XnB: Observation work / ^) // Calculate the coordinate value of the corrected B area _; ^) and the correction amount TB E1C = leg ((C (:, 1), -TC (:, 1) > A2 > ^ ( (C (: 2 ^ 13 1226012 // Calculate the error amount EZOsumPnC-TCW of the C-point selection coordinate value Cl ~ 5 and the correction point TC)) · 2 > i ~ sum (〇rnC-TC (:; 2)) .A2) // Calculate the amount of error between the coordinate values (XnC, YnC) of the C area and the correction point TC " Calculate the error amount of the D area point selection coordinates D1 ~ 5 and the correction point TD EZD ^ sumPnMD ^ l)) · 8 2) ^ sum ((YnI > TD (:; 2)). A2) 〃 Calculate the amount of error between the coordinate values (XnD, YnD) and the correction point TD E1E = dirty, 1), Qing, 1))八 2 > fsum ((3E (: 2), -TE⑽. 22) // Calculate the amount of error E1 ~ 5 between the selected coordinate value E1 and the correction point 区域 E in area E ^ sunKXXnE-TEiyD / ^ fsum ^ XnE-TE ^ y ^) // Calculate coordinate values ρ (ηΕ, ΥηΕ) and correction after correction in E area te's error amount E1 = E1A + E1B + E1C-HE1D + E1E 〃All selected coordinates and correction points ΊE's error amount E2 = E2A + E2B + E2C + E2D + E2E 〃All calibration values and correction points TE The neural network (Neural Network) based on the relationship between the input signal and the output signal is trained to obtain the weight parameters (Wx, Wy) and bias parameters (b) of each block: the weight parameter of the X coordinate of block A (Wx) = [〇 · 9824 0.0373] Bias parameter (b) = 0.0160; Weight parameter of Y coordinate (Wy) = [0 0] Bias parameter (b) = 〇; Weight parameter of X coordinate of block B ( Wx) = [1.0197-0.0385] bias parameter (b) =-0.0317; 1226012 weight parameter of Y coordinate (Wy) = [-0.0049 0.4940] bias parameter (b) = 0.5061; weight parameter of X coordinate of block C (Wx) = [1.0489 -0.0431] bias parameter ⑻ = -0.0358; weight parameter of Y coordinate (Wy) = [-0.0078 0.8031] bias parameter (b) = 0.4093;

The weight parameter (Wx) of the X coordinate of the D block = [1.0159 -0.0183] bias parameter ⑻ = -0.0047; the weight parameter of the Y coordinate (Wy) = [-0.0114 0.8995] the bias parameter (b) = 0.3099; E area Weight parameter of block X coordinate (Wx) = [0.9833 0.0001]

Bias parameter (b) = d〇181; Weight parameter of Y coordinate (Wy) = [0.0159 0.9442] Bias parameter (b) = 0.2845; Coordinates A1 ~ 5, B1 ~ 5 of the measuring point at 25 points , Cl ~ 5, D1 ~ 5, E1 ~ 5 each is based on its relative block multiplied by the block's weight parameter (Wx, Wy), and then the bias parameter 加 is added to obtain the new 25-point corrected The coordinate data is as follows: Block A (XηΑ1, ΥηΑ1) = [-0.0169 0]; (ΧηΑ2, ΥπΑ2) = [1.0514 0]; 15 1226012 (ΧηΑ3, ΥηΑ3) = [1.9543 0]; (χηΑ4, ΥηΑ4) = [ 3.0112 0]; (ΧηΑ5, ΥηΑ5) = [3.9449 0]; Β block (χπΒ1, ΥηΒ1) = [0.0251 0.9963]; (ΧηΒ2, ΥηΒ2) = [0.9897 1.0153]; (ΧηΒ3, ΥηΒ3) = [1.9481 0.9799] (× ηΒ4, ΥηΒ4) = [3.0326 1.0073]; (× ηΒ5, ΥηΒ5) = [3.8706 0.9632]; Block C (XnCl, YnCl) = [0.0532 1.9867]; (XnC2, YnC2) = [0.9442 2.0216]; (XnC3 , YnC3) = [1.9673 1.9934]; (XnC4, YnC4) = [3.0323 1.9985]; (XnC5, YnC5) = [3.8540 1.9784]; D block (XnDl, YnDl) = [0.0209 2.9886]; (XnD2, YnD2) = [0.9989 3.0352]; (XnD3, YnD3) = [1.9477 2.9640]; (XnD4, YnD4) = [3.0287 3.0106]; (XnD5, YnD5) = [3.8793 2.9437]; E block (XnEl, YnEl) = [-0.0040 4.0392]; (XnE2, YnE2) = [l.〇239 4.0043]; (XnE3, YnE3) = [1.9648 3.8843]; (XnE4, YnE4) = [3.0127 4.0577]; (XnE5, YnE5) = [3.9513 4.1244] 'Add the square of the subtraction of A1 ~ 5 and TA1 ~ 5 at each point of block A, and the error amount of the obtained error is 0.0778, (XnAl ~ 5, YnAl ~ 5) is subtracted from TA1 ~ 5 The sum of the squares of the errors is 0.0082. Adding the squares of the subtraction of B1 ~ 5 and TB1 ~ 5 at each point of block B adds 0.0370, (XnBl ~ 5, YnBl ~ 5) Add to the square of the subtraction of TB1 ~ 5, and the resulting error ★ Wu Chadan is 0.0233. Adding the square of the subtraction of C1 ~ 5 and TC1 ~ 5 of the mother point of block C, the resulting error is 0.0735, (XnCl ~ 5, YnCl ~ 5) and the square of TC5 subtraction, The error amount of the obtained error is 0.0305. Add the square of the subtraction of Dl ~ 5 and TD1 ~ 5 at each point of block D, and the error error amount of the obtained error is Na 5 '(Sound 5, YnD 5) and the square of the subtraction and subtraction. : The amount 16 1226012 is 0.0245. For each m-point of a fine block, TE1 ~ 5 subtracts the squares of addition, and the error amount of the obtained error is _Π, (χΠΕ 卜 5, leg ~ 5) The amount of error is 0.0381. Adding five districts_error cents, uncorrected test error = α dove, total corrected error amount E2 = 0.1246 ′, it can be seen that the corrected point improves 56.5% accuracy. After the correction, the difficult-to-reach panel can be corrected again. When the user clicks on the touch panel, the measured value of the red block is substituted for the weight parameter of the block ( Wx, Wy) and the bias parameter _ to establish a correction formula, and use a neural-like algorithm to calculate the corrected coordinates (Xn, Yn) to improve the accuracy of the touch signal. In addition to the above-mentioned implementation of the financial algorithm calculation-level correction formula, it can also be modified by a neural-type algorithm calculation to obtain the money expression. For example, the correction is established as:

Xn ^ WxUX2 + Wxl ^ X + b \

Yn = Wyi ^ Y2 ^ -Wy2 ^ Y + b; This is a two-layer (2 layers) neural algorithm operation, and then uses a neural-like learning method to obtain weight parameters such as Wxl, Wx2, Wyl, Wy2, b, etc. And bias parameters to establish a multi-level correction. In summary, the present invention divides the touch panel into several blocks, and sighs each point in each block, and selects the selected coordinate values obtained by selecting the corrections as input signals. The set correction calibration values are used as the rotation signal, and the correction parameters are established according to the learning rules of the input and output information to determine the weight parameters and bias parameters of the block, and the corrected coordinate values are obtained. In order to improve the accuracy of clicking the touch signal. 17! 226〇12 The disclosure of the above-mentioned spiritual embodiments is used to illustrate the present invention, and is not intended to limit the present invention. Therefore, any change in numerical values or replacement of equivalent components should still belong to the scope of the present invention. From the above detailed description, those skilled in the art can understand that the present invention can indeed achieve the foregoing objectives, and that it has indeed complied with the provisions of the Patent Law, and filed a patent application. [Brief description of the drawings] FIG. 1 is a schematic diagram of the first embodiment of the present invention, which divides the touch panel into 25 blocks and provides 25 correction points. FIG. 2 is a neural network of the present invention or the like. Flowchart of the calibration method. Figure 3 is a schematic diagram of the positions of the selected coordinates, the original coordinates of each calibration point, and the corrected coordinates in the first embodiment of the present invention. Figure 4 is the second embodiment of the present invention to distinguish the touch panel. It is a schematic diagram of five blocks and twenty-five correction points. Fig. 5 is a flowchart of a neural network correction method such as the second embodiment of the present invention. Fig. 6 is a coordinate of each point in the second embodiment of the present invention. Schematic diagram of the original coordinates of each correction point and the coordinates of each corrected point 18

Claims (1)

1226012 Pick up 1. Scope of patent application: A touch panel calibration method, which divides the touch panel into several blocks, and sets at least one calibration point in each block. The selected coordinate value obtained at the right time is used as the input signal, and the original coordinate value of each calibration point is set as the output signal. The relationship between the input signal and the output signal is calculated by a neural network-like learning method. The transfer of alkaloids will be based on the _4 parameter and the touch parameter to establish a correction formula, and the _God's algorithm calculates the correction formula, and calculates the coordinate value after the correction. 2. According to the neural network correction method such as the touch panel described in the scope of the patent application, the touch panel is divided into 25 blocks by a matrix of five by five, and each block is separately Set-correction points to obtain correction formulas for neural-like learning rules for correction. According to the neural network correction method of the touch panel described in item 1 of the application, the touch panel is divided into five blocks according to the Y axis of its coordinates, and each block is based on the X axis of the coordinates. Set the five health positive points to the equal division ②, and obtain the correction formula of five blocks based on the neural learning rule of division and observation. 4 ·, the range of profit conversion; [the method for correcting neural networks such as touch panels as described in [item 4], where 4 is based on the relationship between the input signal and the output signal through the network learning method, and is based on the input The relationship between the signal and the output signal is trained to obtain the observation parameter and the fish bias parameter to establish a multi-layered correction formula. / 19