CN106056339A - Object identification method utilizing penetrability of electromagnetic wave - Google Patents

Abstract

The invention discloses an object identification method utilizing penetrability of electromagnetic waves. The object identification method comprises the steps of adhering RF tags which are common in a logistics system to the bottom part of a box, capturing signals transmitted by the different tags at the adjacent time points, eliminating influence of the ambient environment and an object in motion (such as a walking person or a moving object) in the environment, and acquiring influence of an object in the box on electromagnetic signals. By utilizing the influence, whether the object in the box is illegally stolen or destroyed can be identified. The object identification method provided by the invention is different from a previous method of identifying and checking objects or packets through identifying the authenticity of tags, loads the actual condition and attributes of the object onto the electromagnetic signals, truly realizes the identification of the object, eliminates potential safety hazards in the logistics system, and has better security when compared with the existing method.

Description

A kind of article authentication method utilizing electromagnetic wave penetrance

Technical field

The present invention relates to Internet of Things field of sensing technologies, be specifically related to a kind of article qualification side utilizing electromagnetic wave penetrance Method.

Background technology

In logistics now, storage field, REID the most extensively should use.Compared to two before The dimension technology such as code, bar code, the advantage of REID is: (1), except label can be utilized to carry certain information, is gone back Can utilize reader that information is modified easily.(2) it is not limited to label is placed on appointment orientation, as long as by label extremely In the read range of reader, operation just can be written and read.

In existing radio frequency identification equipment, it is divided into hyperfrequency, high frequency etc., is respectively used under different scenes and application.

In the warehousing system of logistics, tradition differentiates article or the method for parcel, is on the packing box of parcel or article Print Quick Response Code, the mode of bar code.When identifying and identify article, by scanning Quick Response Code and bar code, read therein Article are identified and identify by information.But, potential safety hazard the biggest under such mode: (1) is first, two-dimentional Code and bar code are easy to replicate.Assailant needs only to replicate this Quick Response Code and bar code just can substitute or just pretend to be Normal goods.(2) secondly, scanning Quick Response Code and bar code are a kind of modes differentiating article appended by it, and in logistics and storage In environment, usually printing Quick Response Code or bar code on external packing box, so, assailant is not destroying Quick Response Code and bar code Under premise, the means can disassembled by violence, substitute or take away the article in external packing box.Two kinds of above potential safety hazards, On the premise of mounted box of not unpacking, it is not easy very much to be found and discover.

Summary of the invention

For above-mentioned problems of the prior art, it is an object of the invention to, it is provided that a kind of based on EPC Global The application of the hyperfrequency under Class1Gen2 agreement, it is possible to can differentiate again to wrap up while identifying parcel interior article whether by Replace and the method taken away, it is adaptable to logistics, the situation such as warehousing system.

In order to realize above-mentioned task, the present invention by the following technical solutions:

A kind of article authentication method utilizing electromagnetic wave penetrance, comprises the following steps:

Step one, the data training of article to be measured

Step S10, the bottom that tag array is attached in chest, and article to be measured are put in chest and is packaged；

Step S11, is provided with Wireless RF identifier and general software radio peripheral hardware outside chest, the most wireless penetrates Frequently the read antenna of evaluator is positioned at the lower section of chest, and the antenna of general software radio peripheral hardware is positioned at the top of chest；

Step S12, utilizes the read antenna described in the monitoring of general software radio peripheral hardware antenna and chest interior label array Communication process, obtain primary signal；

Step S13, intercepts out by the EPC signal that tag array in primary signal is replied, and extracts letter from EPC signal Number feature, utilizes grader to determine the label belonging to EPC signal；

Step S14, after the EPC signal alignment of labels different in tag array, removes in external environment condition and environment The impact of mobile article, it is thus achieved that training set, and preserve training set；

Step 2, the qualification of article to be measured

Step S20, time actually detected, for a chest equipped with article to be measured, the bottom in chest has and step S10 The tag array that same procedure is arranged；Chest equipped with article to be measured is carried out by the method identical according to step S11 to step S14 Process, the most now finally give test set rather than training set；

Step S21, compares test set with training set, to determine whether article to be measured are replaced and destroy.

Further, described tag array includes two labels, the respectively first label and the second label.

Further, described grader uses K-means grader.

Further, the detailed process obtaining training data in described step S14 includes:

Remember that the first label, the second label obtain corresponding primary signal S₁、S₂It is respectively as follows:

S₁=F₁+N_en+N_eq+N_w (1)

S₂=F₂+N'_en+N'_eq+N'_w (2)

In above formula, F₁、F₂Represent the first label respectively, the signal time signal of the second label penetrates article to be measured in chest Feature, N_en、N'_enIt is that the external environment condition residing for chest is on the first label, the impact of the second label signal, N_eq、N'_enIt is to use not Same equipment is on the first label, the impact of the second label signal, N_w、N'_wIt it is the Gauss white noise of the first label, the second label signal Sound；

By primary signal S₁、S₂Subtract each other and can obtain:

S₁-S₂=(F₁-F₂)+(N_en-N'_en)+(N_eq-N'_eq)+(N_w-N'_w) (3)

In above formula, the external environment condition residing for chest is similar for the first label and the second label, therefore (N_en-N'_en)≈0；With in once test, the equipment that the first label and the second label use is the same, therefore (N_eq-N'_eq) ≈0；First label, the white Gaussian noise of the second label signal are separate, make G=N_w-N'_w；Then according to white Gaussian noise Probability distribution can obtainI.e. obeying and be desired for 0, variance isDistribution；Make H=F₁-F₂, by Unrelated with article in chest in Gaussian noise, it can be considered that G and H is separate, so the phase is asked on formula (3) both sides Prestige is:

E(S₁-S₂)=E (H)+E (G) (4)

Due to the signal replied for each first label, the second label, up to ten thousand sampled points can be collected, because of This, is by Chebyshev's law of great number:

$\underset{n\→\mathrm{\∞}}{\lim }\Σ\[(N_w-N_w^{\′})\]-\underset{n\→\mathrm{\∞}}{\lim }\ΣE\left(G\right)\≤\mathrm{\ϵ}---\left(5\right)$

In above formula, n represents sampled point；

The variance of known G is 0, therefore can obtain:

$\underset{n\→\mathrm{\∞}}{\lim }\Σ\[S_1-S_2\]=\underset{n\→\mathrm{\∞}}{\lim }\Σ(F_1-F_2)---\left(6\right)$

Therefore, described training set is F₁、F₂Difference constitute eigenmatrix F；In eigenmatrix F, each behavior The characteristic vector that first label, the EPC signal correspondence of the second label obtain after subtracting each other.

Further, after step S15 obtains training set, training set is carried out characteristics extraction, in order to the ratio of step 2 To process；Specific features value extracting method is as follows:

Each row of eigenmatrix F is carried out wavelet transformation, approximate characteristic vector is designated as c_a, minutia vector is remembered For c_d, by the approximate characteristic vector c often gone_aThe new eigenmatrix D constituted_m×k', wherein, m is characterized the number of vector, and k' is Characteristic vector c after wavelet transformation_aLength；

To new eigenmatrix D_m×k'In each row vector extract trend vector V_t, and to new eigenmatrix D_m×k'Whole Body extracts detail matrices V_d, its extract process as shown in Equation 7:

$\begin{array}{c}{V}_t\left(i\right)=\frac{{\&Sigma;}_{j=1}^m{D}_{m\&times;k^{\&prime;}}(j,i)}{m},0<i\&le;k^{\&prime;}\\ V_d(j,i)=D_{m\&times;k^{\&prime;}}(j,i)-V_t\left(i\right),0<i\&le;k^{\&prime;},0<j\&le;m\end{array}---\left(7\right)$

Further, the method in described step S21 compared test set and training set includes:

Step S210, time actually detected, carries out feature according to the method that training set carries out characteristics extraction to test set Value is extracted, and obtains the trend vector V of test set_t' and detail matrices V_d′；

Step S211, calculates training set and the trend vector V of test set_t、V_t' cross-correlation coefficient x；

Step S212, it is judged that cross-correlation coefficient x, whether more than threshold value T, if being more than, carries out step S213, otherwise, it is determined that knot Really R=0, exports R；

Step S213, by training set and detail matrices V of test set_dAnd V_d' put into Naive Bayes Classifier, obtain pre- Survey result R, export R.

The present invention compared with prior art has a techniques below feature:

1. the irregular deformation such as the crack during ultrasonic inspection is primarily used to detection article, and ultrasonic inspection is to thing Impact or damage that product are likely to result in can not be estimated, therefore it is not suitable for logistics, the environment of storage；And the present invention uses electromagnetic wave Penetrance article are identified, can detect parcel in article whether be replaced or take away, further increase logistics, storage Safety detection measure in environment, has great importance to real life and production；

2.X optical check is usually applied and place, the transport hub such as airport, subway, violated in checking the casing of closing Article.But, in a lot of logistics environments, user focuses on individual privacy protection the most very much, it is undesirable that article quilt in chest Stranger checks.Further, owing to delivery side and recipient are often different customer groups, therefore, by checking article in chest In the mode of concrete condition can not confirm chest, whether article are replaced.And due to the constant magnitude of x-ray examination machine, institute Not to be suitable for the environment such as logistics, storage yet.Therefore the present invention can be good at solving the problems referred to above, and identification result is effectively and can Lean on.

Accompanying drawing explanation

Fig. 1 is the flow chart of embodiment in the present invention；

Fig. 2 be in Fig. 1 signal acquisition module schematic diagram is set；

Fig. 3 is the impacts on electromagnetic signal of all kinds of article, and wherein (a) is carton, and (b) is empty van, and (c) is fiber system Product, (d) is spitball (implant)；

Fig. 4 is EPC first kind second filial generation agreement flow chart；

Fig. 5 is K-means classifier result, and wherein (a) is variance and the meansigma methods relation of Alien 964X, and (b) is sound The variance of the outstanding EH47 of frequency and meansigma methods relation, (c) is audio frequency outstanding person's E41C variance and meansigma methods relation, and (d) is Alien 964X's Variance and amplitude relation, (e) is variance and the amplitude relation of audio frequency outstanding person EH47, and (f) is audio frequency outstanding person's E41C variance and amplitude relation；

Fig. 6 is the approximate characteristic vector c of five kinds of unlike materials_aFigure；

In figure, label represents: 1 article to be measured, the antenna of 2 general software radio peripheral hardwares, 3 chests, 4 labels Array, the read antenna of 5 Wireless RF identifiers.

Detailed description of the invention

The Integral Thought of the present invention is, a pair radio-frequency (RF) tag is put into chest (parcel) bottom center, at a distance of 4cm (label Spacing can be according to casing size adjustment, and the chest size used in the present embodiment is 23cm*24cm*40cm), after putting into article, In case top collection training or test data, owing to the dielectric constant of different materials is different, electromagnetic wave is penetrating different article Time suffered impact the most different, by contrastive test collection and the data characteristics of training set, judge to wrap up interior article whether by Replacing or take away, concrete method is as follows:

A kind of article authentication method utilizing electromagnetic wave penetrance, comprises the following steps:

Step one, the data training of article to be measured

Step S10, during training, the bottom that tag array is attached in chest, and article to be measured are put in chest And be packaged；

Step S11, is provided with Wireless RF identifier and general software radio peripheral hardware outside chest, the most wireless penetrates Frequently the read antenna of evaluator is positioned at the lower section of chest, and the antenna of general software radio peripheral hardware is positioned at the top of chest；

Step S12, utilizes the read antenna described in the monitoring of general software radio peripheral hardware antenna and chest interior label array Communication process, obtain primary signal；In this step, general software radio peripheral hardware is same with wireless radio frequency identification reader Time work, obtain the complex signal of article to be measured in label and reader signal penetrate chest；

Step S13, intercepts out by the EPC signal that tag array in primary signal is replied, and extracts letter from EPC signal Number feature, utilizes grader to determine the label belonging to EPC signal；According in EPC Global Class1Gen2 to label and Reader communication modes and the regulation of reader command prefix code, intercept required label signal, owing to label signal is the most weak, bad Identify, therefore by identifying that the signal that reader is sent judges and intercepts the information of tag return.By EPC Global Class1Gen2 agreement understands, before the EPC signal of tag return, reader and the ACK order sent, and return immediately preceding label After multiple EPC signal, it it is the QueryRep signal that sent of reader.Owing to ack signal has fixing with QueryRep signal And it is different from the prefix code of other reader signal, therefore, by identifying and looking for what reader in primary signal was sent Ack signal and QueryRep signal, it is possible to find and intercept the EPC signal of tag return.

The label signal that can obtain due to generic radio peripheral hardware contains many noises, and owing to penetrating casing A large amount of energy losses that middle article are caused, therefore acquired label signal is not easy correct decoding, it is not recommended that with decoding Mode determines the label sending this signal.But the hardware differences caused in manufacturing process due to label, causes each mark Sign the signal replied and there is when backscatter different link frequencies (BLF) and signal characteristic, will by Fourier transformation Have a large amount of noise time-domain signal be converted into frequency-region signal, the trickle difference can being more clearly observed on this link frequency Not.By K-means grader, clearly signal can be divided into two classes.So, can be exactly by K-means grader Judge the label belonging to label signal received in generic radio peripheral hardware.

In the present embodiment, tag array includes two labels, the respectively first label and the second label.

Step S14, after the EPC signal alignment of labels different in tag array, removes in external environment condition and environment The impact of mobile article, it is thus achieved that training set, and preserve training set；

The detailed process obtaining training data includes:

Remember that the first label, the second label obtain corresponding primary signal S₁、S₂It is respectively as follows:

S₁=F₁+N_en+N_eq+N_w (1)

S₂=F₂+N'_en+N'_eq+N'_w (2)

In above formula, F₁、F₂Represent the first label respectively, the signal time signal of the second label penetrates article to be measured in chest Feature, N_en、N'_enBe the external environment condition residing for chest on the first label, the impact of the second label signal, in external environment condition The facility reflection etc. to signal；N_eq、N'_enBeing to use different equipment, such as different antennas, reader, the impact on signal is right First label, the impact of the second label signal；N_w、N'_wIt is the first label, the white Gaussian noise of the second label signal；

By primary signal S₁、S₂Subtract each other and can obtain:

S₁-S₂=(F₁-F₂)+(N_en-N'_en)+(N_eq-N'_eq)+(N_w-N'_w) (3)

In above formula, the external environment condition residing for chest is similar (label for the first label and the second label Spacing is much smaller than the label distance to external environment condition facility, therefore thinks that the signaling reflex of label pair is considered as by external environment condition It is similar), therefore (N_en-N'_en)≈0；Further, with in once test, the equipment that the first label and the second label use is The same, therefore (N_eq-N'_eq)≈0；First label, the white Gaussian noise of the second label signal are separate, make G=N_w- N'_w；Then can obtain according to the probability distribution of white Gaussian noiseI.e. obeying and be desired for 0, variance is Distribution (It is respectively the first label, the second label correspondence white Gaussian noise variance of probability distribution)；Make H=F₁-F₂, Owing to Gaussian noise is unrelated with article in chest, it can be considered that G and H is separate, so formula (3) both sides are asked It is desired for:

E(S₁-S₂)=E (H)+E (G) (4)

Due to the signal replied for each first label, the second label, up to ten thousand sampled points can be collected, because of This, is by Chebyshev's law of great number:

$\underset{n\&RightArrow;\mathrm{\&infin;}}{\lim }\&Sigma;\&lsqb;(N_w-N_w^{\&prime;})\&rsqb;-\underset{n\&RightArrow;\mathrm{\&infin;}}{\lim }\&Sigma;E\left(G\right)\&le;\mathrm{\&epsiv;}---\left(5\right)$

In above formula, n represents sampled point, and ε is positive count；

The variance of known G is 0, therefore can obtain:

$\underset{n\&RightArrow;\mathrm{\&infin;}}{\lim }\&Sigma;\&lsqb;S_1-S_2\&rsqb;=\underset{n\&RightArrow;\mathrm{\&infin;}}{\lim }\&Sigma;(F_1-F_2)---\left(6\right)$

Therefore, when using numerous sampled points to extract feature, the difference of label pair can be regarded as signal characteristic difference Approximation.Described training set is F₁、F₂Difference constitute eigenmatrix F；In eigenmatrix F, each behavior first is marked The characteristic vector that label, the EPC signal correspondence of the second label obtain after subtracting each other.

For the ease of follow-up comparison process, can utilize wavelet transformation that eigenmatrix F is extracted feature further here Value.Wavelet transformation is a kind of new transform analysis method, and it is inherited and has developed the thought of short time discrete Fourier transform localization, with Time overcome again window size not with shortcomings such as frequency changes, using the teaching of the invention it is possible to provide " T/F " window with frequency shift, It is by the ideal tools of signal time frequency analysis and process.It is mainly characterized by by conversion can fully outstanding problem some The feature of aspect, progressively can be entered signal (function) by flexible shift operations the localization analysis of time (space) frequency Row multi-scale refinement, is finally reached high frequency treatment time subdivision, frequency segmentation at low frequency, can automatically adapt to wanting of time frequency signal analysis Ask, thus any details of signal can be focused on.Utilize wavelet transformation, can trend and the fine feature of just signal all extract Come.Specific features value extracting method is as follows:

Each row of eigenmatrix F is carried out wavelet transformation, approximate characteristic vector is designated as c_a, minutia vector is remembered For c_d, it is found through experiments, approximate characteristic vector c_aCan preferably reflect the feature of object, therefore, every to eigenmatrix F After one row vector carries out wavelet transformation, by the approximate characteristic vector c often gone_aThe new eigenmatrix D constituted_m×k', wherein, m Being characterized the number of vector, k' is characteristic vector c after wavelet transformation_aLength；

To new eigenmatrix D_m×k'In each row vector extract trend vector V_tI () (hereafter uses V_tRepresent), and to newly Eigenmatrix D_m×k'Overall extraction detail matrices V_d(j i) (hereafter uses V_dRepresent), i, j are variable parameter；It extracts process such as Shown in formula 7:

$\begin{array}{c}{V}_t\left(i\right)=\frac{{\&Sigma;}_{j=1}^m{D}_{m\&times;k^{\&prime;}}(j,i)}{m},0<i\&le;k^{\&prime;}\\ V_d(j,i)=D_{m\&times;k^{\&prime;}}(j,i)-V_t\left(i\right),0<i\&le;k^{\&prime;},0<j\&le;m\end{array}---\left(7\right)$

Step 2, the qualification of article to be measured

Step S20, time actually detected, for a chest equipped with article to be measured, the bottom in chest has and step S10 The tag array that same procedure is arranged；Chest equipped with article to be measured is carried out by the method identical according to step S11 to step S14 Process, the most now finally give test set rather than training set；I.e. process during this step is the most encapsulated to liking one Good chest, has article to be measured, and tag array is arranged at the bottom in chest in chest.With the method mentioned in step one, Wireless RF identifier and general software radio peripheral hardware are set, make chest positioned there between, it is thus achieved that primary signal, intercept EPC signal, extracts signal characteristic, determines the label described in EPC signal；The EPC signal alignment of different labels is processed, except going Boundary affects thus obtains test set；

Step S21, compares test set with training set, to determine whether article to be measured are replaced and destroy, specifically Method is as follows:

Algorithm inputs:

(1) the trend vector V of chest during the training stage_t；

(2) the trend vector V of chest during test phase_t′；

(3) detail matrices V of chest during the training stage_d；

(4) detail matrices V of chest during test phase_d′；

(5) test threshold T；

Algorithm exports: result of determination R whether parcel is destroyed or replace；

Step S210, time actually detected, carries out feature according to the method that training set carries out characteristics extraction to test set Value is extracted, and obtains the trend vector V of test set_t' and detail matrices V_d′；It is to be the step for of i.e. with test set (replacement training set) Object of study, processes test set according to above-mentioned specific features value extracting method thus obtains V_t' and V_d′；

Step S211, calculates training set and the trend vector V of test set_t、V_t' cross-correlation coefficient x；

Step S212, it is judged that cross-correlation coefficient x, whether more than threshold value T, if being more than, carries out step S213, otherwise, it is determined that knot Really R=0, exports R；

Step S213, by training set and detail matrices V of test set_dAnd V_d' put into Naive Bayes Classifier, obtain pre- Survey result R, export R.

Wherein, test threshold T can be drawn by specific experiment, or is dynamically determined according to the user's request in actual scene. First the trend phasor of twice test result is compared by this algorithm, thinks higher than the test set data of threshold value and is probably normally Data, i.e. wrap up not destroyed and replace.The most directly regarding as improper data less than the test set data of threshold value, i.e. parcel can Destruction can be have passed through or replace.Further detail matrices is compared higher than the test set of threshold value, use naive Bayesian two-value Grader finally determines that package status is the most normal.

Embodiment:

Seeing Fig. 1, the flow process of the present invention is divided into four big modules, respectively signal acquisition module, pretreatment module, feature Extraction module and feature comparing module.Below whole flow process is further elaborated:

1. signal acquisition module

The gatherer process of signal and system are arranged and are seen Fig. 2, are placed in by reader antenna bottom chest, are put by two labels Entering bottom chest, cabinet interior loads institute's shipped item, and general software radio peripheral hardware antenna is placed in case top, collects general The signal of the label reader communication process that software radio peripheral hardware is collected is as the input of pretreatment module.

Seeing Fig. 3, the present invention is at the signal label EPC signal discovery collected by signal acquisition module in intercepting, is passing through In chest after different article, electromagnetic wave signal can produce different signal characteristics, be in particular on signal be signal decay, The formation of burr.

2. pretreatment module

(1) the EPC signal segment of cutting label: the signal collected by general software radio peripheral hardware is whole RFID system System communication process, whole process sees Fig. 4.Therefore firstly the need of intercepting cutting label E PC fragment as signal characteristic abstraction value The source extracted.

(2) utilize grader distinguish label:

Label E PC fragment owing to being syncopated as is the original signal not decoded, therefore firstly the need of distinguishing this segment signal Which by label sent；Owing to signal is the most noisy, easily there is mistake in the mode therefore decoded, therefore utilizes label self Hardware differences carries out judging preferably；This programme is selected K-means grader, the label E PC fragment of intercepting is carried out discrete Result grader after Fourier transformation makes a distinction, and distinguishes result and sees Fig. 5；The experiment of Fig. 5 have employed three class labels, Being respectively Alien 964X, Impinj H47 and Impinj E41, in Fig. 5, (a) (b) (c) is to number the EPC of two labels to set It is set to only one difference, and (d) (e) (f) is that the EPC of two labels is numbered the entirely different of setting.Can by classification results Finding, the method distinguishing label by hardware differences is the most efficient.

(3) align and subtract each other label E PC fragment: at described in detail above elimination ambient noise, equipment diversity, multipath The method of effects, will the EPC signal alignment of two labels and subtracting each other, the signal difference vector of gained extracts eigenvalue.

3. characteristic extracting module

The signal difference of gained in pretreatment module is extracted eigenvalue, and first this programme carries out small echo change to signal difference matrix Change, extract approximate characteristic vector c_a.The approximate characteristic vector c extracted under unlike material_aTool is very different, in order to this is described Point, uses 5 class experiment materials in this programme: metallic article (alloy), fibre (clothes), carton (paperbox), Water (water), woodwork (wood) is respectively put into chest, its approximate characteristic vector c_aSignal as shown in Figure 6, its from trend, The features such as signal intensity have obvious difference.

By approximate characteristic vector c_aConstitute new eigenmatrix D, new feature matrix D is extracted two category feature values, respectively It is trend vector V_tWith detail matrices V_d, this two category feature is left in data base, to treat follow-up comparison.

4., when actually detected, extract training set and the trend vector sum detail matrices of test set, utilize in step S21 and carry To algorithm carry out Characteristic Contrast, the article that i.e. be can determine that in chest by comparing result whether be not hacked person malice replace.

Claims (6)

1. the article authentication method utilizing electromagnetic wave penetrance, it is characterised in that comprise the following steps:

Step one, the data training of article to be measured

Step S10, the bottom that tag array is attached in chest, and article to be measured are put in chest and is packaged；

Step S11, is provided with Wireless RF identifier and general software radio peripheral hardware outside chest, and wherein less radio-frequency is known The read antenna of other device is positioned at the lower section of chest, and the antenna of general software radio peripheral hardware is positioned at the top of chest；

Step S12, utilize general software radio peripheral hardware antenna monitor described in read antenna and chest interior label array logical News process, obtains primary signal；

Step S13, intercepts out by the EPC signal that tag array in primary signal is replied, and it is special to extract signal from EPC signal Levy, utilize grader to determine the label belonging to EPC signal；

Step S14, after the EPC signal alignment of labels different in tag array, removes in external environment condition and environment and moves The impact of article, it is thus achieved that training set, and preserve training set；

Step 2, the qualification of article to be measured

Step S20, time actually detected, for a chest equipped with article to be measured, the bottom in chest has identical with step S10 The tag array that method is arranged；According to the identical method of step S11 to step S14 to the chest equipped with article to be measured at Reason, the most now finally gives test set rather than training set；

Step S21, compares test set with training set, to determine whether article to be measured are replaced and destroy.

Utilize the article authentication method of electromagnetic wave penetrance the most as claimed in claim 1, it is characterised in that described label battle array Row include two labels, the respectively first label and the second label.

Utilize the article authentication method of electromagnetic wave penetrance the most as claimed in claim 1, it is characterised in that described grader Use K-means grader.

Utilize the article authentication method of electromagnetic wave penetrance the most as claimed in claim 2, it is characterised in that described step The detailed process obtaining training data in S14 includes:

Remember that the first label, the second label obtain corresponding primary signal S₁、S₂It is respectively as follows:

S₁=F₁+N_en+N_eq+N_w (1)

S₂=F₂+N'_en+N'_eq+N'_w (2)

In above formula, F₁、F₂Represent the first label respectively, the signal characteristic time signal of the second label penetrates article to be measured in chest, N_en、N'_enIt is that the external environment condition residing for chest is on the first label, the impact of the second label signal, N_eq、N'_enIt is to use different setting Standby on the first label, the impact of the second label signal, N_w、N'_wIt is the first label, the white Gaussian noise of the second label signal；

By primary signal S₁、S₂Subtract each other and can obtain:

S₁-S₂=(F₁-F₂)+(N_en-N'_en)+(N_eq-N'_eq)+(N_w-N'_w) (3) outside in above formula, residing for chest Environment is similar for the first label and the second label, therefore (N_en-N'_en)≈0；In with once testing, the first label The equipment used with the second label is the same, therefore (N_eq-N'_eq)≈0；First label, the Gauss white noise of the second label signal Sound is separate, makes G=N_w-N'_w；Then can obtain according to the probability distribution of white Gaussian noiseI.e. take From being desired for 0, variance isDistribution；Make H=F₁-F₂, owing to Gaussian noise is unrelated with article in chest, the most permissible Think that G and H is separate, so formula (3) both sides asked and be desired for:

E(S₁-S₂)=E (H)+E (G) (4)

Due to the signal replied for each first label, the second label, up to ten thousand sampled points can be collected, therefore, By Chebyshev's law of great number:

In above formula, n represents sampled point；

The variance of known G is 0, therefore can obtain:

Therefore, described training set is F₁、F₂Difference constitute eigenmatrix F；In eigenmatrix F, each behavior first The characteristic vector that label, the EPC signal correspondence of the second label obtain after subtracting each other.

Utilize the article authentication method of electromagnetic wave penetrance the most as claimed in claim 4, it is characterised in that step S15 obtains After training set, training set is carried out characteristics extraction, in order to the comparison process of step 2；Specific features value extracting method is such as Under:

Each row of eigenmatrix F is carried out wavelet transformation, approximate characteristic vector is designated as c_a, minutia vector is designated as c_d, By the approximate characteristic vector c often gone_aThe new eigenmatrix D constituted_m×k', wherein, m is characterized the number of vector, and k' is small echo Characteristic vector c after conversion_aLength；

To new eigenmatrix D_m×k'In each row vector extract trend vector V_t, and to new eigenmatrix D_m×k'Entirety carries Take detail matrices V_d, its extract process as shown in Equation 7:

V_d(j, i)=D_m×k′(j,i)-V_t(i), 0 ＜ i≤k', 0 ＜ j≤m.

Utilize the article authentication method of electromagnetic wave penetrance the most as claimed in claim 5, it is characterised in that described step The method in S21 compared test set and training set includes:

Step S210, time actually detected, carry out eigenvalue according to the method that training set carries out characteristics extraction to test set and carries Take, obtain the trend vector V of test set_t' and detail matrices V_d′；

Step S211, calculates training set and the trend vector V of test set_t、V_t' cross-correlation coefficient x；

Step S212, it is judged that cross-correlation coefficient x, whether more than threshold value T, if being more than, carries out step S213, otherwise, it is determined that result R =0, export R；

Step S213, by training set and detail matrices V of test set_dAnd V_d' put into Naive Bayes Classifier, obtain prediction knot Really R, exports R.