外文翻译

Application of Wave Atoms Decomposition and Extreme Learning Machine for Fingerprint Classification

Abdul A. Mohammed, Q.M. Jonathan Wu, and Maher A. Sid-Ahmed

Department of Electrical Engineering,

University of Windsor, Ontario, Canada

{mohammea,jwu,ahmed}@uwindsor.ca

Abstract. Law enforcement, border security and forensic applications are some

of the areas where fingerprint classification plays an important role. A new

technique based on wave atoms decomposition and bidirectional two-

dimensional principal component analysis (B2DPCA) using extreme learning

machine (ELM) for fast and accurate fingerprint image classification is pro-

posed. The foremost contribution of this paper is application of two dimen-

sional wave atoms decomposition on original fingerprint images to obtain

sparse and efficient coefficients. Secondly, distinctive feature sets are extracted

through dimensionality reduction using B2DPCA. ELM eliminates limitations

of classical training paradigm; trains data at a considerably faster speed due to

its simplified structure and efficient processing. Our algorithm combines opti-

mization of B2DPCA and the speed of ELM to obtain a superior and efficient

algorithm for fingerprint classification. Experimental results on twelve distinct

fingerprint datasets validate the superiority of our proposed method.

1 Introduction

Biometric verification has received considerable attention during the last decade due to increased demand for automatic person categorization. Automated classification of an individual based on fingerprints is preferred since they are less vulnerable to be copied, stolen and lost [1] and due to their uniqueness and stability [2,3]. Fingerprint detection is a technology that has been widely adopted for personal identification in many areas such as criminal investigation, access control and internet authentication.

Fingerprint classification algorithms are classified into two categories; local and global. Local feature based methods include correlation, minutiae and ridge feature based matching algorithms. Global features are obtained using mathematical trans-forms; a classifier compares energies of different fingerprints and classifies them based on the global trends. Correlation-based techniques utilize gray level information of an image and take into account all dimensional attributes of a fingerprint, thereby providing enough image resolution. These techniques have been successfully applied for fingerprint classification [5] but they suffer from higher computational cost. Minu-tiae based techniques [7] extract minutiae from two fingerprints and store them as sets of points in a two dimensional plane and execute matching by generating an alignment between the template and the input minutiae set that result in maximum A. Campilho and M. Kamel (Eds.): ICIAR 2010, Part II, LNCS 6112, pp. 246–255, 2010.

? Springer-Verlag Berlin Heidelberg 2010

Application of Wave Atoms Decomposition and ELM for Fingerprint Classification 247 pairings. In low quality fingerprint images minutiae extraction becomes extremely difficult and thus ridge patterns [4] are reliably extracted for classification.

Researchers have also used fast Fourier transforms (FFT) and multi-resolution analysis tools that extract global features from fingerprint images for classification. Fitz and Green [8] used a hexagonal fast Fourier transform (FFT) to transform finger-print images into frequency domain and employed a ‘‘wedge-ring detector’’ to extract features. A fingerprint classifier based on wavelet transform and probabilistic neural network is proposed in [9]. Wilson et al. [10] developed a Federal Bureau of Investi-gation (FBI) fingerprint classification standard that incorporates a massively parallel neural network structure. Other neural network classification schemes, using self or-ganizing feature map, fuzzy neural networks, radial basis function neural network (RBFNN) and ellipsoidal basis function neural networks (EBFNN) have also been proposed [11].

In this work, we present a fast and accurate fingerprint classification algorithm that extracts sparse fingerprint representation using wave atoms decomposition; these co-efficients are dimensionally reduced using bidirectional two-dimensional principal component analysis (B2DPCA). An extreme learning machine (ELM) classifier, based on a fast single hidden layer feedforward neural network (SLFNN), is trained and tested using dimensionally reduced extracted features. The proposed classifica-tion algorithm requires less human interventions and can run at thousand folds faster learning speed than conventional neural networks. ELM determines network parame-ters analytically, avoids trivial human intervention and makes it efficient for online applications.

The remainder of the paper is divided into 5 sections. Section 2 discusses wave at-oms decomposition, followed by a discussion of B2DPCA in section 3. ELM algo-rithm for classification is discussed in section 4 and the proposed method is described in section 5. Experimental results are discussed in section 6.

2 Wave Atoms Decomposition

Wave atoms [12] are the most recent mathematical transforms for harmonic computa-tional analysis. They are a variant of 2D wavelet packets that retain an isotropic as-pect ratio. Wave atoms encompass a sharp frequency localization that cannot be achieved using a filter bank based on wavelet packets and offer a significantly sparser expansion for oscillatory functions than wavelets, curvelets and Gabor atoms. Wave atoms capture coherence of a pattern across and along oscillations whereas curvelets capture coherence only along the oscillations. Wave atoms precisely interpolate be-tween Gabor atoms and directional wavelets in the sense that the period of oscillations of each wave packet (wavelength) is related to the size of essential support via para-bolic scaling i.e. wavelength is directly proportional to square of the diameter.

Two distinct parameters α; indexing multiscale nature, and β representing direc-tional selectivity are adequate for indexing all known forms of wave packet architec-tures namely wavelets, Gabor, ridgelets, curvelets and wave atoms. The triangle formed by wavelets, curvelets and wave atoms, as shown in the Fig. 1, indicates the wave packet families for which sparsity is preserved under transformation. Wave at-oms are defined for α=β=1/2, where α indexes the multiscale nature of the transform,

248 A.A. Mohammed, Q.M. Jonathan Wu, and M.A. Sid-Ahmed

from α = 0 (uniform) to α = 1 (dyadic). β measures the wave packet’s directional se-lectivity (0 and 1 indicate best and poor selectivity respectively). Wave atoms repre-sent a class of wavelet packets where directionality is sacrificed at the expense of preserving sparsity of oscillatory patterns under smooth diffeomorphisms. Essential support of wave packet in space (left) and in frequency (right) is shown in Fig. 2.

Fig. 1. Comparison of different wave packets architectures with respect to multiscale nature and directional selectivity [12]

2.1 1D Discrete Wave Atoms

Wave atoms are constructed from tensor products of adequately chosen 1D wave

packets. Let )

(,x j n m ψrepresent a one-dimensional family of wave packets, where ,0,≥m j and n ∈Z , centered in frequency around ,2,m j

m j πω±=± with j j C m C 2221≤≤ and centered in space around n x j n j ?=2,. Dyadic scaled and translated versions of 0?m

ψare combined in the frequency domain and the basis function is written as:

)2(2)2()(02/,n x n x x j m j j j m j n m ?=?=?ψψψ. (1)

Fig. 2. Wave atoms tiling in space and frequency [12]

Application of Wave Atoms Decomposition and ELM for Fingerprint Classification 249 The coefficients n m j c ,,, for each wave n m j w ,,, are obtained as decimated convolution

at scale 2-j . Input sample u is discretized at xk=kh, h=1/N, k=1,…N , and discrete coef-

ficients D

n

m j c ,, are computed using a reduced inverse FFT inside an interval of size 2j+1π centered about origin:

. )2(? )2(? )2/2:12/2(2 2p 2,,∑∑+?=Ζ∈++=

?j j j k j j j m nk i D

n m j p k u p k e c ππψ (2)

A simple wrapping technique is used for the implementation of discrete wavelet packets and the steps involved are:

1. Perform an FFT of size N on the samples of u(k).

2. For each pair (j,m), wrap the product u

j m ??ψ by periodicity inside the interval [-2j π, 2j π] and perform an inverse FFT of size 2j to obtain D

n

m j c ,,. 3. Repeat step 2 for all pairs (j,m).

2.2 2D Discrete Wave Atoms

A two-dimensional orthonormal basis function with 4 bumps in frequency plane is formed by individually taking products of 1D wave packets. 2D wave atoms are in-dexed by μ=(j,m,n), where m=(m 1,m 2) and n=(n 1,n 2). Construction is not a simple tensor product since there is only one scale subscript j . This is similar to the non-standard or multi-resolution analysis wavelet bases where the point is to enforce same scale in both directions in order to retain an isotropic aspect ratio. In 2D eq. (1) is modified accordingly.

. )2( )2(),(22211121n x n x x x j j m j j m ??+??=ψψ?μ (3)

Combination of (3) and its Hilbert transform provides basis functions with two bumps in the frequency plane, symmetric with respect to the origin and thus directional wave packets oscillate in a single direction.

2 ,2)2()1(?+?+?=+=μμμμ

μμ?????? (4)

)1(μ?and )2(μ?together form the wave atoms frame and are jointly denoted by μ?.

Wave atoms algorithm is based on the apparent generalization of the 1D wrapping strategy to two dimensions.

3 Bidirectional Two Dimensional Principal Component Analysis Principal Component Analysis (PCA) is a data representation technique widely used in pattern recognition and compression schemes. In the past researchers used PCA and bunch graph matching techniques for enhanced representation of face images. PCA cannot capture even a simple variance unless it is explicitly accounted in the training data. In [13] Yang et al. proposed two dimensional PCA for image representation. As

250 A.A. Mohammed, Q.M. Jonathan Wu, and M.A. Sid-Ahmed

opposed to PCA, 2DPCA is based on 2D image matrices rather than 1D vector so the image matrix does not need to be vectorized prior to feature extraction. Instead an im-age covariance matrix is computed directly using the original image matrices.

Let X denote a q dimensional unitary column vector. To project a p x q image ma-trix A to X ; linear transformation Y=AX is used which results in a p dimensional projected vector Y . The total scatter of the projected samples is characterized by the trace of the covariance matrix i.e. matrix of the projected feature vectors, j(X)=tr(S x ), where tr( ) represents the trace of S x , and S x denotes covariance matrix of the pro-jected features.

.])][()[())())(((T T x X EA A X EA A E Y E Y Y E Y E S ??=??=

(5) .)]()([)(X EA A EA A E X S tr T T x ??= (6)

)]()[(EA A EA A E G T t ??= is q x q nonnegative image covariance matrix. If there are M training samples, the j th image sample is denoted by p x q matrix A j .

()(). 1

1∑=??=M j j T j t A A A A

M G (7)

,)(X G X X J t T =

(8) where A represents an average image of all the training samples. Above criterion is called the generalized total scatter criterion. The unitary vector X that maximizes the criterion j(X) is called the optimal projection axes. An optimal value represents a col-lection of d orthonormal eigen vectors X 1,X 2,….X d of G t corresponding to d largest eigen values. A limitation of 2DPCA based dimension reduction is the processing of higher number of coefficients since it works along row directions only. Zhang and Zhou [14] proposed (2D)2 PCA based on the assumption that training sample images are zero mean, and image covariance matrix can be computed from the outer product of row/column vectors of images.

4 Extreme Learning Machine

Feedforward neural networks (FNNs) are widely used in classification techniques due to their approximation capabilities for non-linear mappings. Slow learning speed of FNNs is a major bottleneck encountered, since input weights and hidden layer biases are updated using a parameter tuning approach such as gradient descent algorithm. Huang et al. [16] proposed an extremely fast learning algorithm called ELM for train-ing a SLFNN. ELM randomly assigns input weights and hidden layer biases if the hid-den layer activation function is infinitely differentiable. In ELM a learning paradigm is converted to a simple linear system whose output weights are analytically determined through a generalized inverse operation of the hidden layer weight matrices.

An N dimension random distinct sample (x i ,t i ) where x i =[x i1,x i2,….x in ]T ∈n ? and t i =[t i1,t i2,….t im ]T ∈m ? , ELM with L hidden nodes and an activation function g(x) is modeled as:

Application of Wave Atoms Decomposition and ELM for Fingerprint Classification 251

()(){} , ,...2,1 ,.11N j o b x w g x g L i j i j i i i L i j i

i ==+=∑∑==ββ (9)

Where w i =[w i1,w i2,….w in ]T , βi =[ βi1, βi2,.…βiL ]T represent weight vectors connecting input nodes to an i th hidden node and from the i th hidden node to all output nodes. b i indicates threshold for i th hidden node whereas w i .x j represents an inner product of w i and x j . An ELM can reliably approximate N samples with zero error.

(){}. ,...2,1 ,.1N j t b x w g L i j i j i

i i ==+∑=β (10)

Eq. (10) is modified as H β=T , H=(w 1,…,w L ,b 1,…,b L ,x 1,…,x N ), such that i th column of H is the output of i th hidden node with respect to inputs x 1,x 2,….x N . If the activation function g(x) is infinitely differentiable, it is proved that the number of hidden nodes are such that L<

(). 211T H E t b x w g E N j L i j i j i i ?==???????+=∑∑==ββ (11)

H is determined using gradient descent and the weights w i , βi and bias parameters b i are tuned iteratively with a learning rate ρ. A small value of ρ causes the learning al-gorithm to converge slowly whereas a higher rate leads to instability and divergence to local minima. To avoid these limitations, ELM incorporates a minimum norm least-square solution, and instead of tuning the entire network parameters a random allocation of input weights and hidden layer biases help to analytically determine the hidden layer output matrix H and curtail the problem to a least-square solution of H β=T . H is a non-square matrix, the norm least-square solution of the above linear system becomes β=H*T , where H * is the moore-penrose generalized inverse of H . The above relationship holds for a non-square matrix H whereas the solution is straightforward for N=L . An infinitely small training error is achieved using the above model since it represents a least-square solution of the linear system.

. min *T H T T HH T H ?==?=?∧βββ (12)

5 Proposed Fingerprint Classification Algorithm

The proposed classification scheme is independent of fingerprint patterns and is based on individual features and the number of trained fingerprint classes. Table 1 consists of detailed steps that demonstrate our proposed technique. Our system classifies fin-gerprint images into one of the trained classes; therefore, only one verification process is required per image. Our proposed scheme deals with classification of fingerprint images using ELM design and utilizes dimensionally reduced feature vectors obtained from wave atoms decomposition. Wave atoms decomposition is used for sparse repre-sentation of fingerprint images since they belong to a category of images that oscillate smoothly in varying directions. Discrete 2D wave atoms decomposition is applied on the original fingerprint image to efficiently capture coherence patterns along and across the oscillations. Fingerprint images are digitized using 256 gray levels there-fore a transformation in color space is not required. Dimension of fingerprint images

252 A.A. Mohammed, Q.M. Jonathan Wu, and M.A. Sid-Ahmed

within each database were reduced to 64×64 prior to wave atoms decomposition. Im-age resizing was the only pre-processing performed on all datasets to minimize com-putations and to guarantee uniformity with other methods used for comparison. An orthonormal basis is used instead of a tight frame since each basis function oscillates in two distinct directions instead of one. This orthobasis variant property is important in applications where redundancy is undesired.

In addition to the aforementioned alterations there were no further changes made to the images as it may lead to image degradation. We randomly divide image database into two sets namely training set and testing set. All images within each database have the same dimension, i.e. R ×C . Similar image sizes support the assembly of equal sized wave atoms coefficients and feature vector extraction with identical level of global content. 2D wave atoms decomposition of every image is computed and coefficients are saved as initial feature matrix. Wave atoms decomposition is a relatively new technique for multiresolution analysis that offers significantly sparser expansion, for oscillatory functions, than other fixed standard representations like wavelets, curvelets and Gabor atoms.

Table 1. Outline of our Proposed Classification Scheme INPUT:Randomly divide image database into two subsets TR i and TE j where i={1,2,…,n} and j={1,2,…,m} representing training and test image sets respectively.

OUTPUT: Classifier - f(x)

1.Resize fingerprint images from all databases to R x C .

https://www.sodocs.net/doc/b211674138.html,pute the wave atoms decomposition of each training and test images and extract

feature sets. Each feature set is of dimension R x C . (Refer to section 2 for details of

wave atoms decomposition)

3.Calculate image covariance matrix of test and train images to obtain intermediate

featue matrix.

| n i R i T R i tR A A A A n G 1

)()(1| m j E j T E j tE A A A A m G 1

)()(14.Evaluate the maximizing criteria J(X) for train and test images.

R

tR T R R X G X X J )(E

tE T E E X G X X J )(5.Repeat steps 3-4 on the transposed intermediate feature matrix to obtain B2DPCA

based feature vectors, f p of size U x V .

6.Train Extreme Learning Machine (ELM) classifier: Generate set of 2DPCA based

feature vectors (vectorized feature vectors obtained in previous step) for training.

7.

Classify images with test feature vectors using ELM trained in step 6. Application of ELM based classification on original wave atoms coefficients is computationally expensive due to higher dimensionality of data originating from large image datasets. Outliers and irrelevant image points being included into classification task can also affect the performance of our algorithm; hence B2DPCA is employed to reduce dimensionality of initial feature vectors. Features are extracted by computing 2DPCA of initial feature matrix along image rows, called as intermediate feature ma-trix. 2DPCA is again applied on the transposed intermediate feature matrix along its

Application of Wave Atoms Decomposition and ELM for Fingerprint Classification 253 rows so as to generate a final feature matrix. Application of 2DPCA using the modi-fied approach retains better structure and correlation information amongst neighbor-ing pixel coefficients. Dimensionally reduced wave atoms coefficients are vectorized into a U×V dimension vector, final feature vector, where U×V << R×C.

B2DPCA based feature vectors better retain the global structure of input space and facilitate accurate classification with lower computational complexity, diminished outliers and irrelevant information. ELM is trained using labeled B2DPCA feature vectors and classified using the trained network.

6 Results and Discussion

Extensive experiments were performed using 3 standard and distinctive collections of fingerprint datasets; FVC2000, FVC2002 and FVC2004 [2] to test the practicality of our proposed method. Each dataset consists of four diverse databases generated using various fingerprint acquisition techniques. Each database contains 8 fingerprints of each of the 100 distinctive subjects.

All images were resized to 64×64 in our experiments and 5 images from each data-base were used as prototypes and the remaining 3 for testing to ensure consistency with other methods used for comparison. Experiments were also performed on origi-nal fingerprint image without resizing and consistently better results were obtained since detailed fingerprint information is incorporated at the expense of large feature vectors. Both the testing and training sets of images are decomposed using 2D wave atoms transform using an orthonormal basis function and dimensionally reduced through application of B2DPCA. Dimensionally reduced features are vectorized and classification is performed by using ELM. The above process was repeated 10 times for all the databases and averaged results of few experiments are documented in the paper. The classification accuracy for Db1 database from FVC2000, FVC2002 and FVC2004 is compared with wavelet transform (WT) based RBFNN and EBFNN fin-gerprint classification algorithms. Results, obtained with the proposed method (only 6 principal components are used for consistency with other methods), are compared with the classification accuracy reported in [11] using WT-2DPCA-RBFNN and WT-2DPCA-EBFNN.

Table 2. Fingerprint classification rates (%) for different techniques

Database WT-2DPCA-

RBFNN WT-2DPCA-

EBFNN

Proposed

Method

FVC 2000 91 91 93.25

FVC 2002 87 87 92.63

FVC 2004 86.5 87 89.62 We conclude from the results in Table 2 that our proposed fingerprint classification algorithm performs significantly better than the wavelet based RBFNN and EBFNN fingerprint classification algorithms. In addition to the improved classification accu-racy, our proposed ELM based classifier performs training and testing thousands folds faster than conventional neural network based classification algorithms [15].

254 A.A. Mohammed, Q.M. Jonathan Wu, and M.A. Sid-Ahmed

From Fig. 3 it is evident that several factors influence classification accuracy, namely, fingerprint acquisition techniques, climatic and environmental conditions and most notably the number of principal components. Dataset Db4 from each of the da-tabases is generated using a synthetic fingerprint generator; consequently the effects of environment and other irrepressible conditions are trifling and are substantiated by improved classification accuracy at low principal components.

(a) Db1 (b) Db2

(c) Db3 (d) Db4

Fig. 3. Classification accuracy vs. number of principal components

7 Conclusion

An original supervised fingerprint classification algorithm for multiclass categoriza-tion based on wave atoms decomposition and bidirectional two-dimensional principal component is proposed. Improvements in classification accuracy validate the fact that wave atoms multiresolution analysis offers significantly sparser expansion, for oscil-latory functions, than other fixed standard representations like wavelets, curvelets and Gabor atoms. The proposed classifier is capable of handling marginal fingerprint orientations, illumination variations, moderate pressure changes against the sensor surface and climatic conditions. The algorithm combines the strengths of both B2DPCA and ELM; creates distinctive and improved feature set, an efficient and fast algorithm for fingerprint classification. The proposed fingerprint classification algo-rithm is independent of the number of prototypes used for training and or testing and is also free of the amount of hidden neurons used for classification, unlike traditional

Application of Wave Atoms Decomposition and ELM for Fingerprint Classification 255

neighborhood based classifiers whose accuracy is greatly affected by the number of prototypes and neighborhood size. Law enforcement, multimedia, and data mining related applications can benefit from our proposed classification scheme.

References

1.Jain, A.K., Ross, A., Pankanti, S.: Biometrics: A tool for information security. IEEE

Transaction on Information Forensics Security 1(2), 125–143 (2006)

2.Maltoni, D., Maio, D., Jain, A.K., Prabhakar, S.: Handbook of fingerprint recognition.

Springer, New York (2003)

3.Yager, N., Amin, A.: Fingerprint classification: a review. Pattern Analysis and Applica-

tions 7(1), 77–93 (2004)

4.Jain, A., Chen, Y., Demitrius, M.: Pores and Ridges: Fingerprint matching using Level 3

features. In: Proc. 18th International Conference on Pattern Recognition, vol. 4, pp. 477–480 (2006)

5.Maio, D., Maltoni, D., Cappelli, R., Wayman, J., Jain, A.K.: FVC 2004: Third fingerprint

verification competition. In: Zhang, D., Jain, A.K. (eds.) ICBA 2004. LNCS, vol. 3072, pp. 1–7. Springer, Heidelberg (2004)

6.Bazen, A., Gerez, S.: Systematic methods for the computation of the direction fields and

singular points of fingerprints. IEEE Transaction on Pattern Analysis and Machine Intelli-gence 24(7), 905–919 (2002)

7.Bhanu, B., Tan, X.: Fingerprint indexing based on novel features of minutiae triplets. IEEE

Transactions on Pattern Analysis and Machine Intelligence 25(5), 616–622 (2003)

8.Fitz, A., Green, R.: Fingerprint classification using a hexagonal fast Fourier transform.

Pattern Recognition 29(10), 1587–1597 (1996)

9.Seokwon, L., Boohee, N.: Fingerprint Recognition using Wavelet Transform and Probabil-

istic Neural Network. In: Proc. of International Joint Conference on Neural Networks, vol. 5, pp. 3276–3279 (1999)

10.Wilson, C., Candela, G., Grother, P., Watson, C., Wilkinson, R.: Massively parallel neural

network fingerprint classification system. National Institute of Standards and Technology;

NISTIR 4880 (1992)

11.Luo, J., Lin, S., Lei, M., Ni, J.: Application of dimensionality reduction analysis to finger-

print recognition. In: ISCID, vol. 2, pp. 102–105 (2008)

12.Demanet, L., Ying, L.: Wave Atoms and Sparsity of Oscillatory Patterns. Applied and

Computational Harmonic Analysis 23(3), 368–387 (2007)

13.Yang, J., Zhang, D., Frangi, A., Yang, J.: Two-dimensional PCA: a new approach to ap-

pearance based face representation and recognition. IEEE Transaction on Pattern Analysis and Machine Intelligence 26(1), 131–137 (2004)

14.Zhang, D., Zhou, Z.: (2D)2 PCA: Two-directional two-dimensional PCA for efficient face

representation and recognition. Neurocomputing 69(1), 224–231 (2005)

15.Pankanti, S., Prabhakar, S., Jain, A.K.: On the Individuality of Fingerprints. IEEE Transac-

tions on Pattern Analysis and Machine Intelligence 24(8), 1010–1025 (2002)

16.Huang, G., Zhu, Q., Siew, C.: Extreme learning machine: Theory and applications. Neuro-

computing 70(1), 489–501 (2006)

17.Ross, A., Jain, A.K., Reisman, J.: A Hybrid Fingerprint Matcher. Pattern Recogni-

tion 36(7), 1661–1673 (2003)

外文翻译

Load and Ultimate Moment of Prestressed Concrete Action Under Overload-Cracking Load It has been shown that a variation in the external load acting on a prestressed beam results in a change in the location of the pressure line for beams in the elastic range.This is a fundamental principle of prestressed construction.In a normal prestressed beam,this shift in the location of the pressure line continues at a relatively uniform rate,as the external load is increased,to the point where cracks develop in the tension fiber.After the cracking load has been exceeded,the rate of movement in the pressure line decreases as additional load is applied,and a significant increase in the stress in the prestressing tendon and the resultant concrete force begins to take place.This change in the action of the internal moment continues until all movement of the pressure line ceases.The moment caused by loads that are applied thereafter is offset entirely by a corresponding and proportional change in the internal forces,just as in reinforced-concrete construction.This fact,that the load in the elastic range and the plastic range is carried by actions that are fundamentally different,is very significant and renders strength computations essential for all designs in order to ensure that adequate safety factors exist.This is true even though the stresses in the elastic range may conform to a recognized elastic design criterion. It should be noted that the load deflection curve is close to a straight line up to the cracking load and that the curve becomes progressively more curved as the load is increased above the cracking load.The curvature of the load-deflection curve for loads over the cracking load is due to the change in the basic internal resisting moment action that counteracts the applied loads,as described above,as well as to plastic strains that begin to take place in the steel and the concrete when stressed to high levels. In some structures it may be essential that the flexural members remain crack free even under significant overloads.This may be due to the structures’being exposed to exceptionally corrosive atmospheres during their useful life.In designing prestressed members to be used in special structures of this type,it may be necessary to compute the load that causes cracking of the tensile flange,in order to ensure that adequate safety against cracking is provided by the design.The computation of the moment that will cause cracking is also necessary to ensure compliance with some design criteria. Many tests have demonstrated that the load-deflection curves of prestressed beams are approximately linear up to and slightly in excess of the load that causes the first cracks in the tensile flange.(The linearity is a function of the rate at which the load is applied.)For this reason,normal elastic-design relationships can be used in computing the cracking load by simply determining the load that results in a net tensile stress in the tensile flange(prestress minus the effects of the applied loads)that is equal to the tensile strength of the concrete.It is customary to assume that the flexural tensile strength of the concrete is equal to the modulus of rupture of the

网络营销外文翻译

E---MARKETING (From:E--Marketing by Judy Strauss,Adel El--Ansary,Raymond Frost---3rd ed.1999 by Pearson Education pp .G4-G25.) As the growth of https://www.sodocs.net/doc/b211674138.html, shows, some marketing principles never change.Markets always welcome an innovative new product, even in a crowded field of competitors ,as long as it provides customer value.Also,Google`s success shows that customers trust good brands and that well-crafted marketing mix strategies can be effective in helping newcomers enter crowded markets. Nevertheless, organizations are scrambling to determine how they can use information technology profitably and to understand what technology means for their business strategies. Marketers want to know which of their time-ested concepts will be enhanced by the Internet, databases,wireless mobile devices, and other technologies. The rapid growth of the Internet and subsequent bursting of the dot-com bubble has marketers wondering,"What next?" This article attempts to answer these questions through careful and systematic examination of successful e-mar-keting strategies in light of proven traditional marketing practices. （Sales Promotion;E--Marketing；Internet；Strategic Planning ） 1.What is E--Marketing E--Marketing is the application of a broad range of information technologies for: Transforming marketing strategies to create more customer value through more effective segmentation ,and positioning strategies；More efficiently planning and executing the conception, distribution promotion,and pricing of goods,services,and ideas;andCreating exchanges that satisfy individual consumer and organizational customers` objectives. This definition sounds a lot like the definition of traditional marketing. Another way to view it is that e-marketing is the result of information technology applied to traditional marketing. E-marketing affects traditional marketing in two ways. First,it increases efficiency in traditional marketing strategies.The transformation results in new business models that add customer value and/or increase company profitability.

交通灯外文翻译 2

当今时代是一个自动化时代，交通灯控制等很多行业的设备都与计算机密切相关。因此，一个好的交通灯控制系统，将给道路拥挤，违章控制等方面给予技术革新。随着大规模集成电路及计算机技术的迅速发展，以及人工智能在控制技术方面的广泛运用，智能设备有了很大的发展，是现代科技发展的主流方向。本文介绍了一个智能交通的系统的设计。该智能交通灯控制系统可以实现的功能有：对某市区的四个主要交通路口进行控制：个路口有固定的工作周期，并且在道路拥挤时中控制中心能改变其周期：对路口违章的机动车能够即时拍照，并提取车牌号。在世界范围内，一个以微电子技术，计算机和通信技术为先导的，一信息技术和信息产业为中心的信息革命方兴未艾。而计算机技术怎样与实际应用更有效的结合并有效的发挥其作用是科学界最热门的话题，也是当今计算机应用中空前活跃的领域。本文主要从单片机的应用上来实现十字路口交通灯智能化的管理，用以控制过往车辆的正常运作。 研究交通的目的是为了优化运输，人流以及货流。由于道路使用者的不断增加，现有资源和基础设施有限，智能交通控制将成为一个非常重要的课题。但是，智能交通控制的应用还存在局限性。例如避免交通拥堵被认为是对环境和经济都有利的，但改善交通流也可能导致需求增加。交通仿真有几个不同的模型。在研究中，我们着重于微观模型，该模型能模仿单独车辆的行为，从而模仿动态的车辆组。 由于低效率的交通控制，汽车在城市交通中都经历过长时间的行进。采用先进的传感器和智能优化算法来优化交通灯控制系统，将会是非常有益的。优化交通灯开关，增加道路容量和流量，可以防止交通堵塞，交通信号灯控制是一个复杂的优化问题和几种智能算法的融合，如模糊逻辑，进化算法，和聚类算法已经在使用，试图解决这一问题，本文提出一种基于多代理聚类算法控制交通信号灯。 在我们的方法中，聚类算法与道路使用者的价值函数是用来确定每个交通灯的最优决策的，这项决定是基于所有道路使用者站在交通路口累积投票，通过估计每辆车的好处(或收益)来确定绿灯时间增益值与总时间是有差异的，它希望在它往返的时候等待，如果灯是红色，或者灯是绿色。等待，直到车辆到达目的地，通过有聚类算法的基础设施，最后经过监测车的监测。 我们对自己的聚类算法模型和其它使用绿灯模拟器的系统做了比较。绿灯模拟器是一个交通模拟器，监控交通流量统计，如平均等待时间，并测试不同的交通灯控制器。结果表明，在拥挤的交通条件下，聚类控制器性能优于其它所有测试的非自适应控制器，我们也测试理论上的平均等待时间，用以选择车辆通过市区的道路，并表明，道路使用者采用合作学习的方法可避免交通瓶颈。 本文安排如下：第2部分叙述如何建立交通模型，预测交通情况和控制交通。第3部分是就相关问题得出结论。第4部分说明了现在正在进一步研究的事实，并介绍了我们的新思想。

建筑类外文文献及中文翻译

forced concrete structure reinforced with an overviewRein Since the reform and opening up, with the national economy's rapid and sustained development of a reinforced concrete structure built, reinforced with the development of technology has been great. Therefore, to promote the use of advanced technology reinforced connecting to improve project quality and speed up the pace of construction, improve labor productivity, reduce costs, and is of great significance. Reinforced steel bars connecting technologies can be divided into two broad categories linking welding machinery and steel. There are six types of welding steel welding methods, and some apply to the prefabricated plant, and some apply to the construction site, some of both apply. There are three types of machinery commonly used reinforcement linking method primarily applicable to the construction site. Ways has its own characteristics and different application, and in the continuous development and improvement. In actual production, should be based on specific conditions of work, working environment and technical requirements, the choice of suitable methods to achieve the best overall efficiency. 1、steel mechanical link 1.1 radial squeeze link Will be a steel sleeve in two sets to the highly-reinforced Department with superhigh pressure hydraulic equipment (squeeze tongs) along steel sleeve radial squeeze steel casing, in squeezing out tongs squeeze pressure role of a steel sleeve plasticity deformation closely integrated with reinforced through reinforced steel sleeve and Wang Liang's Position will be two solid steel bars linked Characteristic: Connect intensity to be high, performance reliable, can bear high stress draw and pigeonhole the load and tired load repeatedly.

外文翻译

Journal of Industrial Textiles https://www.sodocs.net/doc/b211674138.html,/ Optimization of Parameters for the Production of Needlepunched Nonwoven Geotextiles Amit Rawal, Subhash Anand and Tahir Shah 2008 37: 341Journal of Industrial Textiles DOI: 10.1177/1528083707081594 The online version of this article can be found at: https://www.sodocs.net/doc/b211674138.html,/content/37/4/341 Published by: https://www.sodocs.net/doc/b211674138.html, can be found at:Journal of Industrial TextilesAdditional services and information for https://www.sodocs.net/doc/b211674138.html,/cgi/alertsEmail Alerts: https://www.sodocs.net/doc/b211674138.html,/subscriptionsSubscriptions: https://www.sodocs.net/doc/b211674138.html,/journalsReprints.navReprints: https://www.sodocs.net/doc/b211674138.html,/journalsPermissions.navPermissions: https://www.sodocs.net/doc/b211674138.html,/content/37/4/341.refs.htmlCitations: - Mar 28, 2008Version of Record >>

建筑-外文翻译

外文文献： Risk Analysis of the International Construction Project By: Paul Stanford Kupakuwana Cost Engineering Vol. 51/No. 9 September 2009 ABSTRACT This analysis used a case study methodology to analyse the issues surrounding the partial collapse of the roof of a building housing the headquarters of the Standards Association of Zimbabwe (SAZ). In particular, it examined the prior roles played by the team of construction professionals. The analysis revealed that the SAZ’s traditional construction project was generally characterized by high risk. There was a clear indication of the failure of a contractor and architects in preventing and/or mitigating potential construction problems as alleged by the plaintiff. It was reasonable to conclude that between them the defects should have been detected earlier and rectified in good time before the partial roof failure. It appeared justified for the plaintiff to have brought a negligence claim against both the contractor and the architects. The risk analysis facilitated, through its multi-dimensional approach to a critical examination of a construction problem, the identification of an effective risk management strategy for future construction projects. It further served to emphasize the point that clients are becoming more demanding, more discerning, and less willing to accept risk without recompense. Clients do not want surprise, and are more likely to engage in litigation when things go wrong. KEY WORDS:Arbitration, claims, construction, contracts, litigation, project and risk The structural design of the reinforced concrete elements was done by consulting engineers Knight Piesold (KP). Quantity surveying services were provided by Hawkins, Leshnick & Bath (HLB). The contract was awarded to Central African Building Corporation (CABCO) who was also responsible for the provision of a specialist roof structure using patented “gang nail” roof

外文翻译中文版(完整版)

毕业论文外文文献翻译 毕业设计（论文）题目关于企业内部环境绩效审计的研究翻译题目最高审计机关的环境审计活动 学院会计学院 专业会计学 姓名张军芳 班级09020615 学号09027927 指导教师何瑞雄

最高审计机关的环境审计活动 1最高审计机关越来越多的活跃在环境审计领域。特别是1993-1996年期间，工作组已检测到环境审计活动坚定的数量增长。首先，越来越多的最高审计机关已经活跃在这个领域。其次是积极的最高审计机关，甚至变得更加活跃：他们分配较大部分的审计资源给这类工作，同时出版更多环保审计报告。表1显示了平均数字。然而，这里是机构间差异较大。例如，环境报告的数量变化，每个审计机关从1到36份报告不等。 1996-1999年期间，结果是不那么容易诠释。第一，活跃在环境审计领域的最高审计机关数量并没有太大变化。“活性基团”的组成没有保持相同的：一些最高审计机关进入，而其他最高审计机关离开了团队。环境审计花费的时间量略有增加。二，但是，审计报告数量略有下降，1996年和1999年之间。这些数字可能反映了从量到质的转变。这个信号解释了在过去三年从规律性审计到绩效审计的转变（1994-1996年，20％的规律性审计和44％绩效审计;1997-1999：16％规律性审计和绩效审计54％）。在一般情况下，绩效审计需要更多的资源。我们必须认识到审计的范围可能急剧变化。在将来，再将来开发一些其他方式去测算人们工作量而不是计算通过花费的时间和发表的报告会是很有趣的。 在2000年，有62个响应了最高审计机关并向工作组提供了更详细的关于他们自1997年以来公布的工作信息。在1997-1999年，这62个最高审计机关公布的560个环境审计报告。当然，这些报告反映了一个庞大的身躯，可用于其他机构的经验。环境审计报告的参考书目可在网站上的最高审计机关国际组织的工作组看到。这里这个信息是用来给最高审计机关的审计工作的内容更多一些洞察。 自1997年以来，少数环境审计是规律性审计（560篇报告中有87篇，占16％）。大多数审计绩效审计（560篇报告中有304篇，占54％），或组合的规律性和绩效审计（560篇报告中有169篇，占30％）。如前文所述，绩效审计是一个广泛的概念。在实践中，绩效审计往往集中于环保计划的实施（560篇报告中有264篇，占47％），符合国家环保法律，法规的，由政府部门，部委和/或其他机构的任务给访问（560篇报告中有212篇，占38％）。此外，审计经常被列入政府的环境管理系统（560篇报告中有156篇，占28％）。下面的元素得到了关注审计报告：影响或影响现有的国家环境计划非环保项目对环境的影响;环境政策;由政府遵守国际义务和承诺的10％至20％。许多绩效审计包括以上提到的要素之一。 1本文译自：S. Van Leeuwen.(2004).’’Developments in Environmental Auditing by Supreme Audit Institutions’’ Environmental Management Vol. 33, No. 2, pp. 163–1721

营销-外文翻译

外文翻译 原文 Marketing Material Source：Marketing Management Author：Philip Kotler Marketing Channels To reach a target market, the marketer uses three kinds of marketing channels. Communication channels deliver messages to and receive messages from target buyers. They include newspapers, magazines, radio, television, mail, telephone, billboards, posters, fliers, CDs, audiotapes, and the Internet. Beyond these, communications are conveyed by facial expressions and clothing, the look of retail stores, and many other media. Marketers are increasingly adding dialogue channels (e-mail and toll-free numbers) to counterbalance the more normal monologue channels (such as ads). The marketer uses distribution channels to display or deliver the physical product or service to the buyer or user. There are physical distribution channels and service distribution channels, which include warehouses, transportation vehicles, and various trade channels such as distributors, wholesalers, and retailers. The marketer also uses selling channels to effect transactions with potential buyers. Selling channels include not only the distributors and retailers but also the banks and insurance companies that facilitate transactions. Marketers clearly face a design problem in choosing the best mix of communication, distribution, and selling channels for their offerings. Supply Chain Whereas marketing channels connect the marketer to the target buyers, the supply chain describes a longer channel stretching from raw materials to components to final products that are carried to final buyers. For example, the supply chain for women’s purses starts with hides, tanning operations, cutting operations, manufacturing, and the marketing channels that bring products to customers. This supply chain represents a value delivery system. Each company captures only a certain percentage of the total value generated by the supply chain. When a company acquires competitors or moves upstream or downstream, its aim is

智能交通灯控制系统_英文翻译

英文 Because of the rapid development of our economy resulting in the car number of large and medium-sized cities surged and the urban traffic, is facing serious test, leading to the traffic problem increasingly serious, its basically are behaved as follows: traffic accident frequency, to the human life safety enormous threat, Traffic congestion, resulting in serious travel time increases, energy consumption increase; Air pollution and noise pollution degree of deepening, etc. Daily traffic jams become people commonplace and had to endure. In this context, in combination with the actual situation of urban road traffic, develop truly suitable for our own characteristics of intelligent signal control system has become the main task. Preface In practical application at home and abroad, according to the actual traffic signal control application inspection, planar independent intersection signal control basic using set cycle, much time set cycle, half induction, whole sensor etc in several ways. The former two control mode is completely based on planar intersection always traffic flow data of statistical investigation, due to traffic flow the existence of variable sex and randomicity, the two methods have traffic efficiency is low, the scheme, the defects of aging and half inductive and all the inductive the two methods are in the former two ways based on increased vehicle detector and according to the information provided to adjust cycle is long and green letter of vehicle, it than random arrived adaptability bigger, can make vehicles in the parking cord before as few parking, achieve traffic flowing effect In modern industrial production,current,voltage,temperature, pressure, and flow rate, velocity, and switch quantity are common mainly controlled parameter. For example: in metallurgical industry, chemical production, power engineering, the papermaking industry, machinery and food processing and so on many domains, people need to transport the orderly control. By single chip microcomputer to control of traffic, not only has the convenient control, configuration simple and flexible wait for an advantage, but also can greatly improve the technical index by control quantity, thus greatly improve product quality and quantity. Therefore, the monolithic integrated circuit to the traffic light control problem is an industrial production we often encounter problems. In the course of industrial production, there are many industries have lots of traffic equipment, in the current system, most of the traffic control signal is accomplished by relays, but relays response time is long, sensitivity low, long-term after use, fault opportunity increases greatly, and adopts single-chip microcomputer control, the accuracy of far greater than relays, short response time, software reliability, not because working time reduced its performance sake, compared with, this solution has the high feasibility. About AT89C51 (1)function characteristics description: AT89C51 is a low power consumption, high performance CMOS8 bit micro-controller, has the 8K in system programmable Flash memory. Use high-density Atmel company the beltpassword nonvolatile storage technology and manufacturing, and industrial 80S51 product instructions and pin fully compatible. Chip Flash allow program memory in system programmable, also suitable for conventional programmer. In a single chip, have dexterous 8 bits CPU and in system programmable Flash, make AT89C51 for many embedded control application system provides the high flexible, super efficient solution. AT89C51 has the following standard function: 8k bytes Flash, 256 bytes RAM, 32-bit I/O mouth line, the watchdog timer, two data pointer, three 16 timer/counter, a 6 vector level 2 interrupt structure, full-duplex serial port, piece inside crystals timely clock circuit. In addition, AT89C51 can drop to 0Hz static logic operation, support two software can choose power saving mode. Idle mode, the CPU to stop working, allowing the RAM, timer/counter, serial ports, interruption continue to work. Power lost protection mode, RAM content being saved, has been frozen, microcontroller all work stop, until the next interruption or hardware reset so far. As shown in