Because the traditional super resolution restoration algorithm edge preserving ability is insufficient and the ringing effect, we propose the projection on convex sets super resolution restoration algorithm based on modified point spread function. First detect the edge of the reference image; Then, modify the traditional point spread function using a weight factor, and the point spread function is divided into the eight directions of 0°,22. 5°,45°,67. 5°,90°,112. 5°,135°,157. 5°, to achieve the effect of reducing the range of point spread function in the edge section; Finally, modify the reference frame iteratively using the improved point spread function, until the error between the estimated gray value and the actual gray value is small to a certain range or the set iteration number has been reached; Exit iteration and get super-resolution restoration image. The quality of restored image is evaluated by the peak signal to noise ratio, mean square error and structural similarity. Experiment results indicate that the peak signal to noise ratio of the two types of test images is improved by3. 46 ~6. 91 dB, the mean square error is reduced by 43. 47 ~87. 82, and the structural similarity is improved by 0. 050 8 ~0. 381 7. It is concluded that the proposed algorithm improves the edge preserving ability of super resolution reconstruction, and improves the quality of the restored image.

In order to solve the problem that super-resolution restoration quality of the power line infrared image is affected by the camera shaking, a new super-resolution method combining SIFT (Scale Invariant Feature Transform)feature matching is proposed. Taking four adjacent frames for example, an image is selected as the reference frame, and the transformation that map the reference frame to the other images is found by SIFT algorithm, which includes feature point detection and registration, removing mismatching pixels by RANSAC(Random Sample Consensus) algorithm and image resampling. Then, take the reference frame and three resampling images into super-resolution reconstruction. The experimental results show that our algorithm can obtain a better super-resolution image compared with conventional super-resolution method.

 To overcome the problem of energy waste caused by tide effect in mobile network, we propose a base station energy-saving mechanism based on power control in heterogeneous cellular network. The proposed scheme formulates a non-cooperative game, where BSs (Base Stations) act as players and energy efficiency act as utility function. Then the mechanism obtains a suboptimal solution by applying PSO(Particle Swarm Optimization) algorithm. The suboptimal solution is approximate to Nash equilibrium of the game model, and then approximate optimal power control results are obtained. Simulation results show the proposed mechanism enhances the energy efficiency of both MBSs (Macro Base Stations) and FBSs(Femto Base Stations).

 In order to make the security system has the capability of early warning of crime, to achieve the real meaning of intelligent. Crime degree theory of internet speech is put forward, it refers to an individual‘’s possibility of crime, which is shown by social network speech. The theory used multiple analytical methods such as naive bayes and SVM (Support Vector Machine), combining criminal psychology, establishing the theoretical framework and mathematical model. Experimental result has shown that it works well.

Random cluster head selection results in energy consuming too fast in LEACH(Low-Energy Adaptive Clustering Hieravchy Protocd), an improved algorithm which can solve this problem named LEACH-E(LEACH based on Energy) is proposed. It selects the cluster head considering the residual energy of every node and the network, makes them a priority to be cluster head for those nodes whose residual energy are higher than the average energy of the network. When normal nodes join the cluster, the join packet contains the energy information, cluster head calculates the cluster average energy and sends it to BS(Base Station) to calculate the network average energy. It is possible to choose BS as cluster head so as to reducing the speed of energy consumption, It proves that network lifetime and throughput have obvious growth, it also has an advantage over the energy efficiency.

For B2D-LDA(Bilateral Two-Dimensional Linear Discriminant Analysis), when the class mean and the global mean are close, it is hard to classify. A new method for extracting discriminant features in dorsal hand vein recognition, termed MB2D-LDA (Modified Bilateral Two-dimensional Linear Discriminant Analysis), is proposed. MB2D-LDA integrates the cluster information in each class by redefining the between-class scatter matrix, if the distance between the class mean and the global mean is close,the distance between the two classes incorporated into the scatter matrix. A nearest neighbor classifier for dorsal-hand vein matching based on Euclidean distance is used. Experimental results show that our presented MB2D-LDA clearly outperforms B2D-LDA, the average recognition rate of MB2D-LDA is 2% higher than that of B2D-LDA without increasing the recognition time.

As the recognition of the emotional category increases in emotion recognition, the number of support vector machine based on SVM (Support Vector Machine) classification increases dramatically. It results in training difficulty, taking up too much memory space and needing too long time. To solve this problem, we propose an emotion recognition algorithm based on hierarchical support vector machine. The algorithm combines the two-dimensional model for emotion theory, based on hierarchical support vector machine, with the help of wavelet decomposition and other technical means to design a hierarchical structure that is more persuasive than the traditional sentiment classification. A complete set of emotion recognition method for pulse signal is constructed. The experimental results show that this method can reduce the number of support vector machine and improve the classification speed under the premise of computing under a high classification accuracy of 43. 5%.

In the positioning system, the traditional SIFT (Scale Invariant Feature Transform) algorithm is
difficult to meet real-time for the corner recognition problem, an improved feature extraction and description
algorithm is put forward, which is called SIFT algorithm of double threshold FAST(Features from Accelerated
Segment Test). This method adopts double threshold FAST for feature extraction, SIFT algorithm for feature
description which can effectively eliminate many non-corner feature points, greatly improve the recognition speed
of the system. According to the thought of cluster and classification, we build a visual landmarks database for
ceiling recognition. The experimental results show that the algorithm can guarantee the correct matching rate and
improve the real-time performance of the system.

 In order to solve the problem that the electric vehicle recovers a little breaking energy, a regenerative
braking energy recovery strategy is designed based on the fuzzy control. In considering the braking characteristics,
reasonable allocation of the front and back brake force friction braking and regenerative braking force, the braking
energy recovery is increased. A simulation model of fuzzy logic control in ADVISOR is proposed. Experimental results
show that the proposed control strategy is relative to the ADVISOR's own recovery strategy, can recycle more braking
energy by 30% or more, effectively solve the less braking energy recovery problem.

 In order to reduce the energy consumption of elevator and improve the operation efficiency of the
elevator, a model based on Bayesian algorithm is proposed. According to the the prior probability and conditional
probability classification, the idea of probability parking at different periods is put forward, and it is combined
with the infrared detection technology to avoid the empty stop of the elevator. In order to verify the actual
operation efficiency of the algorithm, the simulation test is carried out on the elevator model. Based on Bayesian
algorithm the average waiting time is about 47. 8 s, the time compared to the traditional algorithm is improved by
about 28%. The model will treat peak and valley period of the elevator differently, improving the real-time
efficiency of the system.

 In order to detect the blood glucose of diabetic patients quickly and easily, make up for the use of
invasive blood glucose meter which brings pollution and pain and so on, we have designed a noninvasive blood
glucose measurement method, using a 1 650 nm infrared light source to irradiate the human fingertip to receive
light signals and carry on the photoelectric signal conversion, through A/ D(Analogue to-Digital) acquisition and
serial port communication, digital filter and the Fourier transform and so on to complete data processing, finally
store the blood glucose value and output. Verified by the experiment, this method can realize the basic
measurement of blood glucose concentration. It has the characteristics of high speed, no pollution and high
practicability. The blood glucose value can be detected in a noninvasive way, which avoid the pain caused by
invasive detection.

 The traditional LIC (Local Intensity Clustering) model has the characters of slow speed, low precision
and is sensitive to the initial contour position when segment the uneven gray image. The improved LIC model is
proposed and applied to the thyroid nodule segmentation. The local variance variables are introduced and combined
with Gaussian probability distribution to construct the new local gray level description, which effectively improves
the segmentation accuracy. In order to improve the segmentation speed, set the level set evolution in the narrow
band. The segmentation result acquired by marked watershed algorithm is used as the initial contour to overcome
the shortcomings of the traditional model that is sensitive to the initial contour. The experimental results show that
the improved model can realize the fast adaptive segmentation of thyroid nodules, and the accuracy and the
segmentation speed are greatly improved compared to the traditional LIC model.

 The problem that support vector machine based wavelet image coding is difficult to achieve embedded
characteristics was studied. Firstly, a regression tree structure was constructed in wavelet domain to act as the
basic regression data set, which can use the inner-subband and inter-subband correlation simultaneously.
Secondly, a linear dynamic threshold selecting method was put forward to facilitate the successive approximation
and to make the regression data to be harmonious. Thirdly, based on the threshold selected, the SVM (Support
Vector Machine) error parameters ε was dynamically determined in order to achieve multiple regression to the
wavelet coefficients; the significant coefficients could be encoded prior and the impression algorithm was endowed
with embedded characteristics. Finally, the adaptive arithmetic coding method was used to encode the support
vectors and their weights. The experimental results show that, compared with the current similar algorithms, the
PSNR(Peak Signal to Noise Ratio) of the reconstructed image is improved by 1 ~3 dB.

In order to establish models to simulate or approximate the scale-free networks in real world, we
discuss the scale-free feature of cartesian product network which is obtained from conducting cartesian operation
on two scale-free networks. We employ the concept of the smallest dominating set of network model and focus on
building the cartesian product network by conducting the cartesian product operation on the subnetworks of two
smallest dominating sets of two network models. Furthermore we illustrate a specific process of cartesian product
operation, and analyze the topological properties of the resulting cartesian product network. We import two class
of growing tree network models to explain the process of a non-tree network model by doing the cartesian product
operation on the smallest dominating sets of tree network models. We obtain the conclusion that the cartesain
product network obeys power law.

 Term weighting is a weighting process for terms which is based on the term's effect to the
classification. Term weighting is a strategy that assigns weights to terms in order to improve the performance of
text categorization. We propose two new class space density based term weighting scheme, tf-ICSDF and ICSDF-
based. In the experiments, we investigate the effects of the proposed scheme on the RCV1-4 and 20 Newsgroups
datasets using the SVM (Support Vector Machine) as classifiers. The results show that the proposed scheme
outperform other traditional term weighting schemes, such asprob-based, tf-icf and icf-based.

The information is the basis of the securities market. Because the market is different, the degree of
response to information is different, we studied the relationship between Chinese a-share stock market and
information to analyze our country securities market, and puts forward constructive suggestions for China's
securities market. In the past empirical study, scholars at home and abroad studied on how accounting surplus
Information of listed companies, but gave no satisfied and consistent conclusion. Which is based on the new view
of information frequency, studies the nature of the Chinese A-share stock market, proves the existence of delayed
effect which is our country securities A-share market to information ( Information Delaying Effect) and
information curving effect.

In order to resolve the interconnection problem between different equipments produced by different
manufacturers during network transforming and upgrading, by analyzing the network protocol and taking CISCO and
H3C equipment as an example, the possible interconnection problems are summarized, and a method to interconnect
different equipments in a network. Distinguish the different solutions and implementation ways to the same concept
between different manufactures is proposed. Then we configure the equipments according to the international standard
protocol, and find out the private protocol supported by all manufactures to realize the compatibility of equipment by
different manufacturers. In application, we solved the problems of different manufacturers equipment interconnection
during network transforming and upgrading of campus network very well by this method.