Social force model is a classical model in crowd simulation. Since it was proposed in 1995, the model has been widely used and modified. In 2000, the concept of panic degree was added to the model to propose an improved version. Although many studies focus on social force model, there are few studies on this concept. Therefore, some key parameters and the concept of panic degree in the social force model were analyzed, and the change of panic degree was used to explain the phenomenons of "fast is slow" and "herd behavior" in crowd evacuation. To overcome the problem in the original model:very few pedestrians may not follow others or may not evacuate at the exit in some conditions caused by the not detailed enough description of pedestrian perception in social force model, the visual field range description of the pedestrian was added and the self-motion state for the pedestrian was redefined and other methods were performed to optimize the social force model. Experimental results show that the improved model can simulate the crowd's herd phenomenon well and is helpful for understanding the concept of panic degree in social force model.

Aiming at the problems of color distortion and noise in night image dehazing based on Dark Channel Prior (DCP) and atmospheric scattering model method, a Conditional Generated Adversarial Network (CGAN) dehazing algorithm based on Msmall-Patch training (MP-CGAN) was proposed. Firstly, UNet and Densely connected convolutional Network (DenseNet) were combined into a UDNet (U Densely connected convolutional Network) as the generator network structure. Secondly, Msmall-Patch training was performed on the generator and discriminator networks, that was, multiple small penalty regions were extracted by using the Min-Pool or Max-Pool method for the final Patch of the discriminator. These regions were degraded or easily misjudged. And, severe penalty loss was proposed for these regions, that was, multiple maximum loss values in the discriminator output were selected as the loss. Finally, a new composite loss function was proposed by combining the severe loss function, the perceptual loss and the adversarial perceptual loss. On the test set, compared with the Haze Density Prediction Network algorithm (HDP-Net), the proposed algorithm has the PSNR （Peak Signal-to-Noise Ratio） and SSIM （Structural SIMilarity index） increased by 59% and 37% respectively； compared with the super-pixel algorithm, the proposed algorithm has the PSNR and SSIM increased by 59% and 48% respectively. The experimental results show that the proposed algorithm can reduce the noise artifacts generated during the CGAN training process, and improve the night image dehazing quality.

News text usually consists of tens to hundreds of sentences, which has a large number of characters and contains more information that is not relevant to the topic, affecting the classification performance. In view of the problem, a long text classification method combined with attention mechanism was proposed. Firstly, a sentence was represented by a paragraph vector, and then a neural network attention model of paragraph vectors and text categories was constructed to calculate the sentence's attention. Then the sentence was filtered according to its contribution to the category, which value was mean square error of sentence attention vector. Finally, a classifier base on Convolutional Neural Network (CNN) was constructed. The filtered text and the attention matrix were respectively taken as the network input. Max pooling was used for feature filtering. Random dropout was used to reduce over-fitting. Experiments were conducted on data set of Chinese news text classification task, which was one of the shared tasks in Natural Language Processing and Chinese Computing (NLP&CC) 2014. The proposed method achieved 80.39% in terms of accuracy for the filtered text, which length was 82.74% of the text before filtering, yielded an accuracy improvement of considerable 2.1% compared to text before filtering. The emperimental results show that combining with attention mechanism, the proposed method can improve accuracy of long text classification while achieving sentence level information filtering.

In order to solve the image degradation in the non-uniform dust environment with multiple scattering lights, a slices reconstruction method for single image dedusting was proposed. Firstly, the slices along the depth orientation were produced based on McCartney model in dust environment. Secondly, the joint dust detection method was used to detect dust patches in the slices where non-dust areas were reserved but the dust zones were marked as the candidate detected areas of the next slice image. Then, an image was reconstructed by combining these non-dust areas of each slice and the dust zone of the last slice. Finally, a restored image was obtained by a fast guided filter which was applied to the reconstructed area. The experimental results prove that the proposed restoration method can effectively and quickly get rid of dust in the image, and lay the foundation of object detection and recognition work based on computer vision in dust environment.

The purpose of document-level sentiment analysis is to predict users' sentiment expressed in the document. Traditional neural network-based methods rely on unsupervised word vectors. However, the unsupervised word vectors cannot exactly represent the contextual relationship of context and understand the context. Recurrent Neural Network (RNN) generally used to process sentiment analysis problems has complex structure and numerous model parameters. To address the above issues, a Transfer Learning based Hierarchical Attention Neural Network (TLHANN) was proposed. Firstly, an encoder was trained to understand the context with machine translation task for generating hidden vectors. Then, the encoder was transferred to sentiment analysis task by concatenating the hidden vector generated by the encoder with the corresponding unsupervised vector. The contextual relationship of context could be better represented by distributed representation. Finally, a two-level hierarchical network was applied to sentiment analysis task. A simplified RNN unit called Minimal Gate Unit (MGU) was arranged at each level leading to fewer parameters. The attention mechanism was used in the model for extracting important information. The experimental results show that, the accuracy of the proposed algorithm is increased by an avervage of 8.7% and 23.4% compared with the traditional neural network algorithm and Support Vector Machine (SVM).

The reading comprehension of document in Natural Language Processing (NLP) requires the technologies such as representation, understanding and reasoning on the document. Aiming at the choice questions of literature reading comprehension in college entrance examination, a sentence composition model based on the hierarchical composition model was proposed, which could achieve the semantic consistency measure at the sentence level. Firstly, a neural network model was trained by the triple consisted of single word and phrase vector. Then, the sentence vectors were combined by the trained neural network model (two composition methods:the recursion method and the recurrent method) to obtain the distributed vector of sentence. The similarity between sentences was presented by the cosine similarity between the two sentence vectors. In order to verify the proposed method, the 769 simulation materials and 13 Beijing college entrance examination materials (including the source text and the choice question) were collected as the test set. The experimental results show that, compared with the traditional optimal method based on HowNet semantics, the precision of the proposed recurrent method is improved by 7.8 percentage points in college entrance examination materials and 2.7 percentage points in simulation materials respectively.

In order to achieve the real-time rendering of continuous frame for participating media scene, a rendering algorithm based on optical flow was proposed. First, the regional matching method was used to calculate the field of optical flow between key frames. Then the field of optical flow between intermediate frames was calculated by the interpolation method, and the optical coherence function between frames was used to denote the consistency of optical flow to guarantee that the media motion won't be suddenly changed. Finally, the dynamic scene of continuous frames was rendered according to the field of optical flow. In the participating media scene rendering for 5 continuous frames, the efficiency of the proposed algorithm increase nearly five times than that of based on Radial Basis Function (RBF) model, real-time rendering of consecutive frames is implemented and rendering quality is relatively high.

In oral surgery, segmentation of teeth has important application value. However, due to the ambiguity of tooth boundary, the adhesion of adjacent teeth, and the flexible change of topological structure in dental Computer Tomography (CT) images, it is very difficult to achieve the accurate segmentation. To provide a useful reference for researches, this paper explored the search progress of dental CT image segmentation base on level set methods, summarized the traditional methods of dental CT images segmentation, introduced the level set theory briefly, introduced the details of level set methods for teeth segmentation in recent years, studied the energy terms in level set function, and implemented some contrast experiments. In the dental CT images segmentation based on level set method, the energy terms mainly included competitive energy, edge energy, shape prior energy, global intensity prior energy and local intensity energy. The experimental results show that the performance of hybrid model of the level set method is the best. The segmentation accuracies of incisor and molar teeth were 88.92% and 92.34% respectively. Compared to the method of adaptive threshold and level set without re-initialization, the accuracy of hybrid model improved more than 10% overall. With the utilization of image information and prior knowledge, it is expected to improve the accuracy of segmentation by optimizing and innovating the energy term in the level set function.

Aiming at the low data-transmitting efficiency of the traditional message-oriented middleware in power information system, a firm real-time data-transmitting system based on data stream-transmitting mechanism was proposed. Queue caching mechanism was adopted to realize the asynchronous sending and batch confirmation of message. Data stream-transmitting mechanism was designed to eliminate the cache latency and the cost of cache resources of the data on transit node to improve the timeliness and concurrency of data transmission. Distributed and data routing thought was used-data to make the node network to the third-party system transparently and achieve a data routing distribution function. The simulation results of a provincial electric power information system data exchange scene, verified the system performance. Concurrent data exchange capacity is 3000 concurrent. Transmission speed in the gigabit bandwidth system environment is 980 MB/s. Switching delay is kept in milliseconds.

The current medical image registration technology could not meet the real-time requirements for clinical diagnosis and treatment. Graphic Processing Unit (GPU) accelerated medical image registration technology was reviewed and discussed for this problem in this paper. The paper summarized GPU general purpose computation, studied current technology of medical image registration which based on GPU acceleration with the essential framework of medical image registration as main line, and implemented Positron Emission computed Tomography (PET) and Computed Tomography (CT) image registration experiments respectively on Central Processing Unit (CPU) and GPU computing platforms. The Normalized Mutual Information (NMI) value of GPU accelerated medical image registration based on Free Form Deformation ( FFD) and NMI was slightly smaller than that of CPU method, but the registration efficiency is 12 times than CPU method. Except keeping high registration accuracy, GPU accelerated medical image registration algorithms also get a lot of ascension in terms of registration speed.

To improve malodor management and emergency response ability, this paper proposed a regional malodor on-line monitoring platform solution. Referring to network-load equilibrium and dynamic extendibility, the platform implemented real-time monitoring and remote monitoring. The remote monitoring module corresponding to the Remote Desktop Protocol (RDP) could adjust a terminal's parameters, alarm beyond limit and sample at different grades. Based on Advanced Encryption Standard (AES) and MD5 digital-signature technology, a combined algorithm was designed to improve the safety of the RDP. The platform has been piloting in Dagang Petrochemical Industrial Park in Tianjin Binhai New Area，which could accumulate data and experience for the research of future malodor diffusion model as well as malodor pollution control, and make technical preparation for malodor on-line monitoring system to merge into the Internet of Things (IOT) for environmental protection.

Multi-Cell Joint Channel Estimation (MC-JCE) is the practical basis of community collaboration and joint processing technology; however, the MC-JCE needs Power Delay Profile (PDP) knowledge of all channels remains the same and gets known. To solve this problem, based on the cooperative community Orthogonal Frequency Division Multiplexing (OFDM) system framework, this article first explored in the extended multi-cell joint channel estimation (MC-eJCE) algorithm of multi-cell channel PDP differences but known conditions, and then put forward modified multi-cell joint channel estimation (MC-mJCE) algorithm under the condition of this information unknown, at the same time to reduce the computational complexity of algorithm and get minimum Mean Square Error (MSE) of channel estimation. This article used optimal pilot set design of comb pilot to finally derive Cramer-Rao bound of MC-JCE algorithm. The simulation shows the MC-eJCE and MC-mJCE algorithms have good performance of MSE under the conditions of multi-cell channel PDP differences, and Space-Frequency Block Code (SFBC) encoding for joint transmission also has a good performance of Bit Errer Rate (BER) based on the above algorithm for channel estimation.

A method was proposed for estimating building heights via registering catadioptric omni-directional image and remote sensing image, which can be applied to large-scale 3D city reconstruction. Firstly, the top edges of building roof were extracted from the catadioptric omni-directional image by using omnidirectional Hough transform. Then the catadioptric omni-directional image and the remote sensing image were registered based on the extracted top edges where the angle consistency nature of horizontal lines in catadioptric omni-directional imaging was used as evidence. Finally, according to the model of catadioptric omnidirectional camera, the building heights were estimated by using the registration results. The proposed method is simple and easy to implement. The experimental results show that the method is effective and the error of estimated building height is fairly small.

The status of buffer managing methods and the characteristics of the representative workload of DBMS was introduced.By analyzing access patterns,a model of evaluating cost was constructed,and its efficiency under various workloads was validated.The tests show that the adaptive policies based on access patterns of buffering management can get the optimal performance of DBMS.

In the design of DBMS,the high performance of DBMS is important.In this paper,a self-tuning model of dynamic-tinning of resource of DBMS was brought forward based on feedback model.This model can adjust system parameters automatically to achieve good performance.It also puts forward the dialog tree,model of relationship of resource,and the arithmetic used to make tinning plan.