In order to realize accurate management of yaks during the process of yak breeding, it is necessary to recognize the identities of the yaks. Yak face recognition is a feasible method of yak identification. However, the existing yak face recognition algorithms based on neural networks have the problems such as too many features in the yak face dataset and long training time of neural networks. Therefore, based on the method of transfer learning and combined with the Visual Geometry Group (VGG) network and Convolutional Neural Network (CNN), a Parallel CNN (Parallel-CNN) algorithm was proposed to identify the facial information of yaks. Firstly, the existing VGG16 network was used to perform transfer learning to the yak face image data and extract the yaks' facial information features for the first time. Then, the dimensional transformation was performed to the extracted features at different levels, and the processed features were inputted into the parallel-CNN for the secondary feature extraction. Finally, two separated fully connected layers were used to classify the yak face images. Experimental results showed that Parallel-CNN was able to recognize yak faces with different angles, illuminations and poses. On the test dataset with 90 000 yak face images of 300 yaks, the recognition accuracy of the proposed algorithm reached 91.2%. The proposed algorithm can accurately recognize the identities of the yaks, and can help the yak farm to realize the intelligent management of the yaks.

When solving the optimal solutions of 0-1 Knapsack Problems (KPs), the traditional Genetic Algorithm (GA) has insufficient local refinement ability and the simple local search algorithm has limited global exploration ability. Aiming at these problems, two algorithms were integrated to the Hybrid Greedy Genetic Algorithm (HGGA). Under the GA global search framework, local search module was added, and the traditional repair operator based only on item value density was improved, the greedy hybrid option based on item value was added, so as to accelerate the optimization process. In HGGA, the population was led to carry out fine search in the excellent solution space of evolution, and the classical operators of GA were relied on to expand the global search space, so as to achieve a good balance between the refinement ability and the development ability of the algorithm. HGGA was tested on three sets of data. The results show that in the first set of 15 test cases, HGGA is able to find the optimal solution on 12 cases, with a success rate of 80%; on the second small-scale dataset, the performance of HGGA is obviously better than those of other similar GA and other meta-heuristic algorithms; on the third large-scale dataset, HGGA is more stable and efficient than other meta-heuristic algorithms.

Remote sensing image interpretation plays an important role in the acquisition of Land Use and Land Cover (LULC) information, and automatic classification serves as the key to improve the efficiency of LULC information acquisition. The actual scenes have a great mount of inaccurate prior knowledge. Extracting and integrating the available knowledge in the prior knowledge can help to further improve the accuracy, automation rate and scale application ability of image classification methods. Based on the above situation, a new deep learning classification method of Landsat 8 OLI images based on inaccurate prior knowledge was proposed. For the proposed method, inaccurate units in prior knowledge were avoided automatically, realizing automatic region selection and feature extraction of classified samples and obtaining high confidence knowledge in the constraint space of patches. Then, the deep residual network was trained by using these classified samples, and the accurate classification of large-area images was achieved. In the experiment, Xinbei district of Changzhou city was taken as the example, the data of 2009 land use status of this district was selected as the prior data, and the 2014 Landsat 8 OLI image of this district was selected as the to-be-classified image. The experimental results show that the proposed method has advantages such as the integration of inaccurate prior knowledge and the accurate classification of large-area contiguous LULC information. Besides, it can obtain the accurate boundary of main land use patches, and has the accuracy for patch classification in the whole image of 88.7% and the Kappa coefficient of 0.842.The proposed method can cooperate with deep learning method to achieve high precision Landsat 8 OLI remote sensing image classification.

Traditional 2D convolutional models suffer from loss of semantic information and lack of sequential feature expression ability in sentiment classification. Aiming at these problems, a hybrid model based on 1D Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) was proposed. Firstly, 2D convolution was replaced by 1D convolution to retain richer local semantic features. Then, a pooling layer was used to reduce data dimension and the output was put into the recurrent neural network layer to extract sequential information between the features. Finally, softmax layer was used to realize the sentiment classification. The experimental results on multiple standard English datasets show that the proposed model has 1-3 percentage points improvement in classification accuracy compared with traditional statistical method and end-to-end deep learning method. Analysis of each component of network verifies the value of introduction of 1D convolution and recurrent neural network for better classification accuracy.

In order to solve the problem that traditional ensemble algorithms are not suitable for imbalanced data classification, Over Sampling AdaBoost based on Margin theory (MOSBoost) was proposed. Firstly, the margins of original samples were obtained by pre-training. Then, the minority class samples were heuristic duplicated by margin sorting thus forming a new balanced sample set. Finally, the finall ensemble classifier was obtained by the trained AdaBoost with the balanced sample set as the input. In the experiment on UCI dataset, F-measure and G-mean were used to evaluate MOSBoost, AdaBoost, Random OverSampling AdaBoost (ROSBoost) and Random UnderSampling AdaBoost (RDSBoost). The experimental results show that MOSBoost is superior to other three algorithm. Compared with AdaBoost, MOSBoost improves 8.4% and 6.2% respctively under F-measure and G-mean criteria.

Concerning low efficiency and accuracy of the ridesharing user group discovery in shared transportation environment, a GeoOD-Tree index was established based on R-tree principle, and a group discovery strategy to maximize the multiplying rate was proposed. Firstly, the feature extraction and calibration processing of original spatio-temporal trajectory data was carried out to mine effective Origin-Destination (OD) trajectory. Secondly, a data structure termed GeoOD-Tree was established for effective storage management of OD trajectory. Finally, a group discovery model aiming at maximizing ridesharing travel was proposed, and a pruning strategy using by K Nearest Neighbors (KNN) query was introduced to improve the efficiency of group discovery. The proposed method was evaluated with extensive experiments on a real dataset of 12000 taxis in Xi'an, in comparison experiments with Dynamic Time Warping (DTW) algorithm, the accuracy and efficiency of the proposed algorithm was increased by 10.12% and 1500% respectively. The experimental results show that the proposed group discovery strategy can effectively improve the accuracy and efficiency of ridesharing user group discovery, and it can effectively improve the rideshared travel rate.

In view of the problem that conventional incremental learning models may go through a way of performance degradation during the update stage, a kernelized incremental learning method was proposed based on objective equilibrium measurement. By setting the optimization term of "empirical risk minimization", an optimization objective function fulfilling the equilibrium measurement with respect to training data size was designed. The optimal solution was given under the condition of incremental learning training, and a lightweight incremental learning classification model was finally constructed based on the effective selection strategy of new data. Experimental results on a publicly available fall detection dataset show that, when the recognition accuracy of representative methods falls below 60%, the proposed method can still maintain the recognition accuracy more than 95%, while the computational consumption of the model update is only 3 milliseconds. In conclusion, the proposed method contributes to achieving a stable growth of recognition performance as well as efficiently decreasing the time consumptions, which can effectively realize wearable devices based intellectual applications in the cloud service platform.

To reduce the unnecessary overhead of data source authentication in static security mechanism and defend the On-off attack in trust threshold mechanism, an adaptive security mechanism based on trust was proposed in the Internet of Things (IoT). Firstly, the trust evaluation model was built according to node behavior in information interaction, further the measure method for total trust value of nodes was given. Then, for the nodes whose total trust values were higher than the trust threshold, the trust-based adaptive detection algorithm was used to detect the changes of the total trust values of these nodes in real time. Finally, the relay nodes determined whether to authenticate the received message according to the returned result of adaptive detection algorithm. The simulation results and analysis show that the proposed mechanism reduces the energy overhead of relay nodes, and plays a better role in defense against On-off attacks in IoT.

Concerning the problem that the lack of testing methods would lead to mismatch between supply and demand of private clouds, an adaptive and scalable private cloud system testing method was proposed, which can test private cloud computing ability in IaaS (Infrastructure as a Service). The number of virtual machines was dynamically increased through the private cloud application program interface, hardware information and operating system category of the virtual machine configuration were selected by performance-characteristic model, and different load models were used according to different needs of users to form simulation environment. At last, cloud computing Service Level Agreement (SLA) was used as a test standard to measure the ability of private cloud services. The proposed method was implemented in Openstack. The experimental results show that private cloud platform service capacity can be obtained by the proposed method with lower cost and higher efficiency than user test. Compared with Openstack component Rally, scalability and dynamic load simulation of the proposed has greatly been improved.

Since the new generation of flash-based SSD (Solid-State Drivers) use the DDR3 interface as its interface, SSD must communicate with memory controller correctly. FPGA (Field-Programmable Gate Array) was used to design the DDR3 protocol parsing logic. Firstly, the working principle of DDR3 was introduced to understand the controlling mechanism of memory controller. Next, the architecture of this interface parsing logic was designed, and the key technical points, including clock, writing leveling, delay controlling, interface synchronous controlling were designed by FPGA. Last, the validity and feasibility of the proposed design were proved by the modelsim simulation result and board level validation. In terms of performance, through the test of single data, continuous data and mixed read and write data, the bandwidth utilization of DDR3 interface is up to 77.81%. As the test result shows, the design of DDR3 parsing logic can improve the access performance of storage system.

The explosion of data in big data environment brings great challenges to the system structure and capacity of storage system. Nowadays, the development of storage systems tends to be large capacity, low-cost, and high performance. Meanwhile, storage devices such as conventional rotated magnetic Hard Disk Drive (HDD), Solid State Drive (SSD) and Non-Volatile Random Access Memory (NVRAM) have limitations caused by their intrinsic characteristics, leading to the fact that a single kind of storage device cannot meet the requirements above. Hybrid storage which utilized different storage medium was a good solution to this problem. SSD, as a kind of memory wiht high reliability, low energy consumption, high performance is more and more widely applied to hybrid storage system. By combining magnetic disk with the solid-state drives, the advantages of the high performance of SSD and the low-cost, high-capacity features of HDD were taken. The hybrid storage could provide users with large capacity of storage space, guarantee the system's high performance, at the same time reduced the cost. The current research status of the SSD-HDD hybrid storage system was described, different SSD-HDD hybrid storage systems were summarized and classified. In view of two different HDD-SSD hybrid storage architectures, the key technologies and insufficiencies of which were discussed. Prediction of trend and the research focus in the hybrid storage future were discussed at last.

In order to solve the problem that the population diversity preservation strategy only based on crowding distance of Non-dominated Sorting Genetic Algorithm-Ⅱ (NSGA-Ⅱ) cannot reflect the real crowding degree of individuals, an improved NSGA-Ⅱ algorithm based on the adaptive hybrid non-dominated individual sorting strategy (NSGA-Ⅱ ^h) was proposed. First, a novel loop-clustering individual sorting strategy was designed. Second, according to the Pareto layer-sorting information the NSGA-Ⅱ ^h algorithm adaptively chose one from the two individual sorting strategies based on classical crowding distance and loop-clustering. Finally, the diversity maintain mechanism could be improved especially during the late period of evolutionary optimization. The NSGA-Ⅱ ^h algorithm was compared with three classical algorithms including NSGA-Ⅱ, Multi-Objective Particle Swarm Optimization (MOPSO) and GDE3. The experiments on five multi-objective benchmark functions show that the NSGA-Ⅱ ^h algorithm can acquire 80% of optimal Inverted Generational Distance (IGD) values, and the corresponding two-tailed t-test results at a 0.05 level of significance are remarkable. The proposed algorithm can not only improve convergence of the original algorithm, but also enhance the distribution of Pareto optimal set.

In order to improve the decoding quality of depth map video in Distributed Multi-view Video plus Depth (DMVD) coding, a new non-uniform quantization scheme based on the sub-band layer and sub-band coefficients was proposed in wavelet domain Distributed Video Coding (DVC). The main idea was allocating more bits to pixels belong to the edge of depth map and consequently improving the quality of the depth map. According to the distribution characteristics of the wavelet coefficients of depth map, the low frequency wavelet coefficients of layer- N kept the uniform quantization scheme, while the high frequency wavelet coefficients of all layers used the non-uniform quantization scheme. For the high frequency wavelet coefficients around "0", larger quantization step was adopted. As the amplitude of the high frequency wavelet coefficients increased, the quantization step decreased, with finer quantization and the quality of the edge was improved consequently. The experimental results show that, for "Dancer" and "PoznanHall2" depth sequence with more edges, the proposed scheme can achieve up to 1.2 dB in terms of the Rate-Distortion (R-D) performance improvement by improving the quality of edges; for "Newspaper" and "Balloons" depth sequences with less edges, the proposed scheme can still get 0.3 dB of the R-D performance.

As the source of the haze data, the reliability of the haze monitoring sites is very important to the reliability of the big data. Due to the lack of effective evaluation method for the haze monitoring points, the monitoring data is not reliable enough. In order to solve the problem that the perceived data was not reliable, a kind of perceptual source trust evaluation and selection model was proposed based on the data trigger detection method. When the perceived data arrived, the K-Means clustering algorithm and the statistical results were firstly used to calculate the benchmark data, then the trust degree of data was calculated by using the current perceived data, the benchmark data and the threshold values. Secondly, according to the location of the perceptual source, neighbor relationship was determined. The current perceived data and the data of the neighbors were compared, according to the absolute value of the difference and the value of the threshold, the neighbor recommendation trust degree was calculated. Finally, the comprehensive trust degree was calculated by using the truest degree of perceived data, the historical trust degree and the recommendation trust degree of the neighbor. The initial value of the historical trust was set as the number of monitoring items, and then updated by the comprehensive trust. Theoretical analysis and simulation results prove that the proposed method can effectively evaluate the perceived source, avoid the abnormal data, and reduce post processing overhead.

Support Vector Machine (SVM) has good performance in dealing with dimensionality disaster, over fitting and nonlinearity, which other traditional situation assessment algorithms does not have. However SVM has low efficiency when dealing with large-scale data. To effectively confront the challenge of handling big data, a MapReduce-based SVM (MR-SVM) situation assessment algorithm was proposed. Considering the characteristics of SVM algorithm, the parallelization and parameter selection of SVM based on MapReduce programming was implemented by designing procedures of map function and reduce function. The performances of MR-SVM and SVM were compared on Hadoop platform, MR-SVM had lower efficiency than SVM when dealing with small-scale data, but much better performance when dealing with large-scale data. SVM had an exponential growth on training time with the growth of data scalability while MR-SVM has slow growth. The experiment results show that MR-SVM solves the problem of data scalability, therefore it is suitable for situation assessment in big data environment.

Concerning the application requirements for the fast classification of large-scale remote sensing images, a parallel classification method based on K-means algorithm was proposed. Combined the CPU process-level and thread-level parallelism features, reasonable strategies of data granularity decomposition and task scheduling between processes and threads were implemented. This algorithm can achieve satisfactory parallel acceleration while ensuring classification accuracy. The experimental results using large-volume and multi-scale remote sensing images show that: the proposed parallel algorithm can significantly reduce the classification time, get good speedup with the maximum value of 13.83, and obtain good load-balancing. Thus it can solve the remote sensing image classification problems of the large area.

Multiple Directed Acyclic Graphic (DAG) scheduling algorithms are supposed to take many factors into account, such as execution time, communication overhead, cost and fairness of all DAG. Therefore, in order to increase fairness and reduce cost, a new scheduling strategy CAFS (Communication Aware Fair Scheduling), based on CA-DAG (Communication Aware-DAG), was proposed. Also, a BD (Backward Difference) rule was introduced to optimize finish time of all DAGs. CAFS is consisted of two parts: the pre-scheduling part adopts CACO (Communication Aware Cost Optimization) to optimize the total cost, and utilizes the classical fairness algorithm to decide the sequence for scheduling. Based on the sequence the second part schedules all the DAGs using BD rule to reduce the finish time. The simulation results show that CAFS can reduce the finish time without increasing cost and guarantee the fairness, and the average execution time has been reduced by 19.82%.

Methods of parallel computation are used in validating topology of polygons stored in simple feature model. This paper designed and implemented a parallel algorithm of validating topology of polygons stored in simple feature model. The algorithm changed the master-slave strategy based on characteristics of topology validation and generated threads in master processor to implement task parallelism. Running time of computing and writing topology errors was hidden in this way. MPI and PThread were used to achieve the combination of processes and threads. The land use data of 5 cities in Jiangsu, China, was used to check the performance of this algorithm. After testing, this parallel algorithm is able to validate topology of massive polygons stored in simple feature model correctly and efficiently. Compared with master-slave strategy, the speedup of this algorithm increases by 20%.

The mouse behaviors cannot be accurately recognized by the existing computer-based automatic analysis system, and the ground truth is generally achieved from experts’ annotation on a massive number of video images. However, to some extent, subjective misjudgments are unavoidable. To solve these problems, a human computation-based mouse behavior recognition method was proposed in this paper. Because of the superiority of human visual perception, and the decentralization and cooperation of the internet, human brains were treated as processors in a distributed system. Firstly, every mouse behavior frames were distributed to on-line individuals randomly, and each behavior frame was classified by a large number of on-line individuals. Secondly, all the effective classifications from the on-line individuals were collected, analyzed and processed by computer system, realizing the final mouse behavior classification based on these frame sequences. The experimental results show that the proposed method is effective to improve the correct recognition rate of mouse behaviors with limited cost.

The diversity of population, the searching capability and the robustness are three key points to the multi-objective optimization problem, which directly affect the convergence of algorithm and the spread of solutions set. To better deal with above problems, a Scatter Search hybrid Multi-Objective Evolutionary optimization Algorithm (SSMOEA) was proposed. The SSMOEA followed the scatter search structure but designed a new selection strategy of diversity and integrated the method of co-evolution in the process of subset generation. Additionally, a novel adaptive multi-crossover operation was employed to improve the self-adaptability and robustness of the algorithm. The experimental results on twelve standard benchmark problems show that, compared with three state-of-the-art multi-objective optimizers, SPEA2, NSGA-Ⅱ and AbYSS, SSMOEA outperforms the other three algorithms as regards the coverage, uniformity and approximation. Meanwhile, its robustness is also significantly improved.

To resolve the problem of the complex 3D tunnel modeling generated by the changes of the tunnel sections at different rock formations during the project, the 3D tunnel modeling based on the characteristic sections was proposed. Through establishing the characteristic section model library, the 3D modeling of the changed tunnel sections was realized and the efficiency of the modeling of the complex tunnel sections was improved. Following the illustration of the data collection method and an analysis of the characteristic sections with detailed coordinates, the smoothing algorithm (smoothing the tunnel axis at the corner by an arc) of the changed tunnel sections at the corner was proposed. For the ordinary tunnel sections, the triangulation was applied in the 3D modeling; for the complex sections with simple quadrilateral structure, the 3D modeling was realized by using the Bézier surface method and the surface splicing techniques which has been validated through experiments.

This paper proposed a moving target detection method based on the background image set and sparse representation. The method combined the Robust Principal Component Analysis (RPCA) and the image block analysis method based on sparse representation. The authors got a series of background images from a video sequence by RPCA, and combined these background images as the background image set, treating image block as basic unit, and moving target was extracted from input frame by image block analysis method based on sparse representation. The simulation results indicate that when the background illumination mutates, the proposed method can effectively eliminate the impact of environment noise and reduce the false detection rate of target detection.

Concerning the problems of high time complexity and topological inconsistency existing in the multi-resolution representation of curve for progressive transmission, an algorithm of generating multi-resolution curves for progressive transmission over the Internet was proposed in this paper. By using pre-stored vertex deviation to simplify curves and using an optimized monotone chain intersection algorithm to maintain topological consistency, the algorithm can quickly generate topologically consistent multi-resolution curves. The algorithm was used in the experiment of progressive transmission for curve data, and the experimental results show that the multi-resolution curve data maintain topological consistency and the generation time changes linearly with the amount of data. The effectiveness of the algorithm has been verified in the experiment.

Some E-commerce sites cannot deploy independent recommender systems themselves due to limited resources. In order to help these sites deploy recommender systems quickly and conveniently, an E-commerce recommendation platform based on Software-as-a-Service (SaaS) pattern was proposed and implemented. This platform used unified scripts to collect user action information and provided recommendation services through standard interface. It realized a low coupling between platform and E-commerce sites so that the cost of implementation was reduced. The results of online operations show that this platform can help E-commerce sites increase the conversion rate and the volume of orders.

Concerning the specific application environment of embedded microprocessor, the branch prediction technology was researched in this paper, and a new scheme of branch prediction was proposed. Compatible with cache design, jump direction and destination address of branch prediction happened on extended instruction bus. The unexecuted instruction and address pointer were saved for possible recovery after misprediction, which reduced misprediction penalty, simultaneously guaranteed the instruction flow to execute correctly. The study shows this scheme is of little hardware spending, high prediction efficiency and low misprediction penalty.

Concerning the inequality of cluster-head distribution and node energy consumption in Wireless Sensor Network (WSN) cluster routing protocol, a node-energy load-balance clustering algorithm was proposed. Group according to the node energy, dynamically adjust the group number to the node energy reduction, conduct cluster-head election in the group according to the energy focus, and further balance the node energy consumption using cluster-head rotate and multi-hop routing between clusters. The simulation results show that this protocol effectively balances the energy consumption among network nodes and achieves an obvious improvement in network stable period.

To seek the optimal path for the vehicles to take delivery of auto parts under the Milk-run, a modeling idea that each components supplier's spare parts were delivered by the way of circular and batch delivery to make as full use of the vehicle as possible was put forward. The optimizing model of vehicle routing problem was established with the constraints of vehicle cubage, arriving time window, supplier supplying dynamic time window and maximum running distance. After that, a heuristic saving algorithm (or C-W algorithm) was designed to provide a solution to the model. Finally, one example was given to prove the validity of the model and algorithm.

In order to make it more convenient and more effective to add, delete, update and query a data warehouse through QC-tree, this paper suggested how to incrementally maintain and query QC-tree. The implementation maintains and queries QC-tree based on the structure of QC-tree, combining the deep first algorithm and cover of equivalent classes. The implementation involves only the upper bound of equivalent classes and considers whatever may happen to the states of classes, so that it confirms the effectiveness and correctness. Compared with the traditional maintenance and query through data cube, the implementation dramatically cuts down the data amount to be considered, so that it improves the performance of maintenance and query.