научная статья по теме HYPER SAUSAGE NEURON: RECOGNITION OF TRANSGENIC SUGAR-BEET BASED ON TERAHERTZ SPECTROSCOPY Физика

Текст научной статьи на тему «HYPER SAUSAGE NEURON: RECOGNITION OF TRANSGENIC SUGAR-BEET BASED ON TERAHERTZ SPECTROSCOPY»

ОПТИКА И СПЕКТРОСКОПИЯ, 2015, том 118, № 1, с. 182-187

ГЕОМЕТРИЧЕСКАЯ И ПРИКЛАДНАЯ ОПТИКА

y%K 535.8

HYPER SAUSAGE NEURON: RECOGNITION OF TRANSGENIC SUGAR-BEET

BASED ON TERAHERTZ SPECTROSCOPY

© 2015 n Jianjun Liu*, Zhi Li*, **, Fangrong Hu***, Tao Chen***, Yong Du****, *****, and Haitao Xin****

*School of Mechano-Electronic Engineering, Xidian University, Xi'an, Shanxi 710071, PR China **Guilin University of Aerospace Technology, Guangxi, Guilin 541004, PR China ***School of Electronic Engineering and Automation, Guilin University of Electronic Technology, Guilin,

Guangxi 541004, PR China ****School of Information Engineering College, Jimei University, Xiamen, Fujian 361005, PR China *****Xiamen University, Xiamen, Fujian 361005, PR China E-mail: guet.lizhi@gmail.com Received May 5, 2014

This paper presents a novel approach for identification of terahertz (THz) spectral of genetically modified organisms (GMOs) based on Hyper Sausage Neuron (HSN), and THz transmittance spectra of some typical transgenic sugar-beet samples are investigated to demonstrate its feasibility. Principal component analysis (PCA) is applied to extract features of the spectrum data, and instead of the original spectrum data, the feature signals are fed into the HSN pattern recognition, a new multiple weights neural network (MWNN). The experimental result shows that the HSN model not only can correctly classify different types of transgenic sugar-beets, but also can reject identity non similar samples in the same type. The proposed approach provides a new effective method for detection and identification of GMOs by using THz spectroscopy.

DOI: 10.7868/S0030403415010134

1. INTRODUCTION

Terahertz usually refers to electromagnetic wave with the frequency on 0.1 ~ 10 THz (wavelength 30 um ~ ~ 3 mm), and the band between microwave and infrared, which belongs to the far-infrared band. Theoretical studies show that the vibration and rotational energy levels of most biological molecules (DNA, protein) are in the THz band. Thus, it is possible to use THz time-domain spectroscopy (THz-TDS) for the biological detection and identification [1—4].

With the popularization of transgenic technology and transgenic products, the safety inspection and assessment of transgenic food attracts more and more attention. At present, there are several commonly used transgenic testing protocols mainly including PCR and ELISA detection methods [5—10]. Although those methods have the characteristics of high sensitivity, they need the international standards for the detection of transgenic products to identify and the detection of transgenic products is very inconvenient. Therefore, it is necessary to develop new detection methods. The application of near infrared spectroscopy in the transgenic products has been quite extensive: [11] reports the application of near-infrared spectroscopy for the detection and identification of transgenic corn, [12] describes the application of near-infrared spectroscopy in the detection of erucic acid and glu-cosinolate on transgenic rapeseed, Lijuan Xie et al.

[13] presents a visible/near-infrared spectroscopy analysis technique for the diagnosis of transgenic tomato leaf. The near-infrared technology provides a powerful method for the detection of transgenic products. However, visible/infrared spectra may suffer from the computation and optimizing parameters problems. With the rapid development of terahertz technology, terahertz application is more and more widely used in security, but the application of identify transgenic food by using terahertz is still in the initial stage. Hence, there are some important theoretical and practical significance to detect transgenic food by using terahertz spectroscopic.

2. EXPERIMENTAL SAMPLES AND APPARATUS

2.1. Samples

The different transgenic sugar-beet and its parent are all supplied by the Institute of Cotton Research of Chinese Academy ofAgricultural Sciences (ICRCAAS). Before scanning, all samples are taken 20 g and baked 48 h by 50°C temperatures. Take out 200 mg to create into slices with the diameter 1.2 cm and the thickness 1.5 mm from each sample. Each kind of sample amounts to 100 slices. Each sample is repeated 25 times for scanning. Details of the selected transgenic cottons are presented in Table 1.

Table 1. Transgenic sugar-beet material

Sample No. The tested materials Material category Result of gene test Diameter Shape Thickness Samples

GM01 Transgenic sugar beet A* GM + 1.2 cm sheet 1.5 mm 100

GM02 Parent of sugar beet A GM receptor I 1.2 cm sheet 1.5 mm 100

GM03 Transgenic sugar beet B GM + 1.2 cm sheet 1.5 mm 100

GM04 Parent of sugar beet B GM receptor I 1.2 cm sheet 1.5 mm 100

* Some transgenic sugar-beets used for this experiment has not yet been published. For the protection of intellectual property rights, especially, those hidden names of transgenic sugar beet are replaced by A and B.

2.2. Experimental Apparatus

The time-resolved spectroscopy detection system used in this experiment is the same as what is depicted in Ref. [14]. In order to ensure the accuracy of the experiment, Nitrogen is injected until the internal relative humidity is less than 2%. Indoor relative humidity is 25% and temperature is 292 K.

2.3. Terahertz Spectroscopy of Transgenic Sugar-Beet

Figures 1, 2 show the time and frequency domain graph of transgenic sugar-beet after vector normalization is preprocessed.

Figure 3 demonstrates the characteristics absorption peaks of different transgenic sugar-beets. As seen from the Fig. 3, the terahertz spectra of two different transgenic sugar-beets and their parents are very similar. Although different transgenic sugar-beets are implanted into the A and B genes in their respective parent, most of the spectral information of transgenic sugar-beet inherited the characteristics of its parents. In this case, it is difficult to identify directly the samples by their characteristic spectral features. To realize

Amplitude, arb. un.

_i_i

0 10 20

Time, ps

Fig. 1. Time domain spectroscopy of transgenic sugar-beet.

identification, we can adopt pattern recognition methods to identify THz spectra of these samples [15, 16].

It can be seen from Fig. 3 that the THz spectroscopy transgenic sugar-beet and their parents are very similar. We must have the aid of a pattern recognition method to identify transgenic beet and their parents. In view of this, HSN is used to establish identification model for transgenic sugar-beet [17].

3. IDENTIFICATION MODEL OF HYPER SAUSAGE NEURON (HSN)

3.1. Spectral Data Preprocessing

The terahertz spectral information of transgenic sugar-beet is obtained in 0.2—1.5 THz range in this paper. It can be known from the spectral data that the dimension of original spectrum data is very high with the small number of sample. It is not appropriate either from the model performance or the computational complexity to use high dimensional data to establish identification model directly [18, 19]. PCA is a most commonly used method for extracting feature, which linearly combines multiple variables of original data according to the principle of maximum variance and

Fig. 2. Frequency domain spectroscopy of transgenic sugar-beet.

ОПТИКА И СПЕКТРОСКОПИЯ том 118 № 1

2015

184

JIANJUN LIU et al.

Absorbance, arb. un. - GMO1

2

---GMO2

...... GMO3

----GMO4

Y = f [®(X W1W2,..., Wm) - Th],

(1)

where Xis the input vector, W1,W2,..., Wm — are the weight vector, O — represents the function relationship between input and weight vectors Th are the threshold, and f is the discriminant function.

When m = 2, Eg. (1) expressed as two weights neuron, we named it as hyper sausage neuron (HSN). The mathematical model of HSN can be represented as shown in the following:

Y = f [®(XWW) - Th], 0(XW1W2) = X -0 (Wi,WJ ,

(2) (3)

PC2 -1.9

-2.1

-2.3

0.3 0.6 0.9 1.2 1.5

Frequence, THz -2.5

:<r*

Jo o

(A)

D DflB

° nn ° □

□o □

(B)

P

• •

1.4

1.5

1.6

Fig. 3. Absorption peaks of different transgenic sugar-beet.

uses the new fewer variables replace the original high dimensional variable set to achieve dimension reduction.

Figure 4 is the two-dimensional spatial distribution graph of transgenic sugar-beets and their parents by principal component (PC) analysis. It can be seen from Fig. 4 that all samples are zonal distribution in the principal component space and have the obvious dividing line between cultivars, aiming to achieve the objective of classification.

In range A, GM01 is a transgenic sugar-beet A and its parent is GM02, in range B, GM03 and GM04 are transgenic sugar-beet B and its parents, respectively. It is shown in Fig. 4 that the density of space distance of sample can reflect the genetic relationship between samples.

3.2. Structure of Hyper Sausage Neuron

The multiple weights neural network (MWNN) is a more general model of neural network which is proposed by Wang Shoujue[20]. A multi weight neuron can be expressed as

1.7 PC1

Fig. 4. Two-dimensional graph of PC1 vs PC2 of all samples: GMO1 (■), GMO2 (□), GMO3 (•), GMO4 (O).

<W2) = {Y|Y = flWi + (1 - a)W2,a e [0,1]}, (4)

where 0

(W1W2)

represents a one-dimensional line that is determined by W1 and W2 in n-dimensional space, and 0(X ,W1,W2) describes the Euclidean distance from X

to 0(WbW2).

Discriminant function f is described as following:

f (x) =

1, x < 0, -1, x > 0.

(5)

Coverage area of HSN network is actually a topological product of finite one-dimensional line and hyper sphere in «-dimensional space, the radius of hyper sphere is threshold Th.

Choosing different neurons values of the order p and weight W ^, different shapes of HSN may be obtained. Figure 5 lists different shapes of HSN when different value of order p and weight Wy is selected. This paper selectsp = 2, Wn = 1.0, W12 = 1.0 to build model.

3.3. Establishment of Coverage Area in High Dimensional Space Network

Step 1. Set the all network point set of some types of transgenic sug

Для дальнейшего прочтения статьи необходимо приобрести полный текст. Статьи высылаются в формате PDF на указанную при оплате почту. Время доставки составляет менее 10 минут. Стоимость одной статьи — 150 рублей.

Показать целиком