Abstract:In view of the fact that the previous saliency detection models fail to fully consider the effect of stereo visual comfort and the distribution features of disparity values, a saliency computation model considering stereo visual comfort is proposed. In the extraction of color image's saliency, the model first segments an input image into super-pixel regions by using SLIC algorithm, and merges the regions according to color similarity among adjacent regions. After that, the computation of 2D image's saliency is conducted. In the computation of depth saliency, the model first preprocesses the disparity map, and then a regional disparity contrast-based saliency analysis is applied to compute the salient region of the depth map. Finally, the stereo visual comfort factor is embedded into the fusion of the 2D saliency map and depth map to obtain a final stereoscopic saliency image. We evaluated the proposed model for stereoscopic images with various scenarios. The experimental results indicate that the proposed model outperforme existing saliency detection models, yielding an 85% precision and 78% recall rate. Moreover, the saliency region distributions fit well with the human binocular visual attention.

Abstract:Identifying anchor links on social networks plays an important role in cross-network information dissemination, cross-platform recommendation, prediction of social chain, and so on. To improve the accuracy of anchor links identification, this paper proposes an effective method:the IAUE model. Firstly, the model uses network embedding algorithms to draw the network representation based on network structure. Then, a candidate set of matching nodes is gotten by BP neural network, stochastic gradient descent and negative sampling strategies. To refine the result of anchor links match, the G-S algorithm is used to reduce the useless information. Experiments upon multiple data sets show that the IAUE method has better performance and good generalization ability compared with other approaches. This research to some extent also can accurately identify the anchor links in the social network.

Abstract:The industrial Internet has become a representative technology in the fourth industrial revolution. According to the demands of data transmission service in industrial network, such as stable topology and traffic flow changes regularly, an artificial immune strategy-based routing mechanism in industrial cognitive wireless network is proposed to realize the reliable routing of industrial network, which includes the static routing algorithm in intra-domain based on link quality and the dynamic routing algorithm in inter-domain based on multipath. A static routing algorithm in intra-domain based on link quality is proposed, which combines the hardware and software to monitor the network link and calculates the link packet loss rate according to the moving window index weighted average method. A dynamic routing algorithm is proposed in inter-domain based on multipath. According to the model distance of the node to predict the traffic to prevent nodes from losing packet caused by excessive traffic. The simulation results from the OMNET++ simulation platform show that compared with the pair-wise directional geographical routing algorithm, the proposed routing mechanism reduces the packet loss rate and network overhead respectively in response to the burst traffic. Compared with the graph routing algorithm, the packet loss rate is 4 times lower than that in the case of link failure.

Abstract:The mobile social cloud is a new paradigm that combines mobile cloud with social cloud. It can provide users with a safe and reliable resource sharing platform. In the conventional resources allocation, the time overhead is large for mobile users to obtain resources from a remote data center, resulting in significantly degrading quality of users' experience. At the same time, there is little consideration of the trust relationship established by users based on social attribute, which leads to credit risk and low reputation of users in transaction mechanism. Therefore, a novel resources allocation mechanism of mobile social cloud is designed in this paper considering the social and economic benefits of users. Firstly, to promote the sharing of resources among social friends, the improved Gale-Shapley algorithm is used to match appropriate sellers for the buyers in their friend circles. Then, to maximize the use of idle mobile cloud resources, a multi-to-multi buyer bidirectional sealed-bid auction algorithm is used to reallocate resources for the unsuccessfully matching users. Finally, the proposed resources allocation mechanism is simulated. The simulation results show that the proposed resources allocation mechanism in this paper has better performances than the Incentive-Compatible Auction Mechanism in terms of price satisfaction, social reputation satisfaction and success rate of resources transaction.

Abstract:The Low duty cycle wireless sensor networks (called LDC-WSNs) can be deployed in harsh environments that humans are difficult to access to perform long-term monitoring and target tracking tasks, and they have broad application prospects. LDC-WSN reduces the energy consumption caused by idle listening, but its end-to-end delay is great. The existing LDC-WSN routing protocols focuses on the reduction of end-to-end delay, but they do not fully consider the balancing of energy consumption of nodes in the network. Therefore, some nodes would consume their energy quickly and die soon. To solve this problem, this paper considers the link quality and node's energy levels, and then propose a novel routing algorithm EADR (energy-aware dynamic routing). In EADR, each node maintains a forward set, which contains a set of neighbor nodes with high quality links. During the data transmission, a sender node would send its data to its neighbors with high decision-making factors, where the decision-making factors are dynamically determined by node's work/sleep schedules and energy levels. Simulation results show that EADR can reduce the end-to-end delay, achieve higher successful rate of data transmission, and extend the network lifetime.

Abstract:Aiming at the loading imbalance of subdomain controller on multi-domain deployment in SDN, based on genetic migrating population and the ideas of migrating switch, this paper has proposed the load balancing mechanism for multi-domain controller in SDN. It integrates main network overheads and designs the optimal migration domain selecting algorithm according to the biological genetic thought, getting the optimal immigration and emigration domain. At the same time, contrasting with the phenomenon of population migration, this paper designs the competition and migrating algorithm for switch based on survival time and elimination mechanism, which balances the number of switches for each subdomain. Compared with the existing algorithm, the simulation shows that the process of migration domain has been significantly optimized. Moreover, it balances the control overheads in subdomain and effectively ensures the balanced distribution of the controller loads.

Abstract:The short text classification is a key task in the field of Internet text data processing. The long short-term memory (LSTM) and convolutional neural network (CNN) are the two most important deep learning models for short text classification. The research on the deep learning in the field of computer vision and speech recognition shows that the deep level of neural network model has better ability to express data features. Inspired by this, a deep learning model named ResLCNN (residual-LSTM-CNN) is proposed based on the structure of three LSTM layers and one CNN layer for text deep learning classification problem. In this model, the LSTM layer is used to capture long distance dependency features of the sequence data and the CNN layer can extract local features of the sentence by convolution operators. The ResLCNN model combines the advantages of LSTM and CNN effectively. At the same time, based on the residual model theory, the ResLCNN model adds an identity mapping between the first LSTM layer and CNN layer to alleviate the problem of vanishing gradients. In order to explore the ability of ResLCNN model for deep short text classification, some experiments are made on several data sets to compare with LSTM, CNN and their combination models. The result shows that compared with the single LSTM and CNN combination model, the ResLCNN deep model improves the accuracy rate by 1.0%, 0.5% and 0.47% respectively on the data sets of MR, SST-2 and SST-5 and achieves better classification results.

Abstract:To improve the cache performance of content-centric mobile edge networks, a user mobility-aware and node centrality based caching mechanism, UMANCC, is proposed. UMANCC utilizes edge nodes to calculate the node centrality, idle rate of cache and cell sojourn time of users. A mobile edge network controller collects aforementioned information of edge nodes, and ranks each node base on the calculated importance factor of nodes. The caching node is selected according to the ranking result. Simulation results show that compared with the traditional caching mechanisms such as LCE and Prob, UMANCC effectively reduces the average number of hop count of fetching contents by up to 15.9%; it improves the hit ratio of cache at edge nodes by at least 13.7%, and it reduces traffic flowing into core networks by up to 32.1%. It greatly improves the performance of content distribution in content-centric mobile edge networks.

Abstract:The flow scheduling on cloud data center network is a hotspot at present. More practical flow scheduling assume that the flow information is unknown in advance, but the performance of these scheduling scheme is not good enough. By combining the multi-level feedback queue with the per-flow scheduling, this paper designs a multiple level scheduling scheme FISH based on the flow isolation, which solves the queueing competition of different flows. Experimental results show that the proposed scheme has stable performance and reduces the completion time of small flows by up to 8.6%.

Abstract:TLS is the most widely deployed security protocol, however, it can only secure the applications that are based on reliable transport. Datagram TLS (DTLS) is a modified version of the TLS protocol which provides security protection in datagram environments. In DTLS, however, the communication parties need complete authentication though the certification authority when they establish connection. Consequently, the connection establishment process takes long time with a high security overhead, which cannot meet the requirement for resource-constrained network communication environment such as Internet of Things. This paper introduces identify-based cryptography to DTLS. It provides authentication while calculating the session key, and avoids the overhead associated with handling certificates in the handshake protocol. The paper designs a new DTLS handshake protocol, which reduces the number of interactions and messages, and shortens the connection establishment time. Experimental results show that the DTLS with identity-based cryptography reduces the communication setup time by nearly 50% without compromising the security.