Cloud Computing and Security

In response to the increasing food safety problems of the meat product in the industry chain of livestock and poultry, and the actual demand of enterprises and consumers, combining the development status of the enterprises, considering the safety factors of animal breeding and product processing, this paper designed a food safety traceability system, based on modern internet of things technologies such as RFID, QR code, sensor technology and wireless communication technology, of which the traceability information covers all links of the whole industry chain. Compared with other traceability systems, the system in this paper has improved greatly in terms of its breadth, depth, and accuracy, which is an effective way to solve the food safety problem. The system provides the customers, enterprises, and the government department with multi-channel information traceability service, which has the real significance of the reduction food safety risk and improvement of recall efficiency of defective products.

In this paper, we propose a secure key pairing system, FaceMatch, to enable confidential communication for smartphones. The idea is to leverage a natural human interaction mode: face-to-face interaction. Two users raise up their smartphones to establish a secure channel based on their similar observations via the smartphones’ cameras. One user acts as the initiator and uses the front-facing camera to snap his/her face. The other user captures the initiator’s face using his/her rear camera. Utilizing the facial appearance as a secret substitution, the initiator delivers a randomly generated key to the other user. Based on this key, two users establish a confidential communication channel between their devices. FaceMatch achieves secure device pairing without the complex operations needed in prior works. We implemented FaceMatch using off-the-shelf iPhones. The experimental results show that FaceMatch can establish a 128-bit encrypted connection in less than 3 s.

Software Defined Networking (SDN) is a new type of network architecture, which provides an important way to implement automated network deployment and flexible management. However, security problems in SDN are also inevitable in industrial networks. In the research area of SDN security and traditional network security, feasibility and influence of defense in depth in industrial networks should thus be explored. In this paper, a security classification transmission method based on SDN in industrial networks is proposed, which provides a better security level of transmission paths. In the proposed method, the security classification transmission system is first presented. By designing five service mechanisms, including request, strategy generation, distribution/maintenance, updating/loading and execution, the security classification transmission service model is defined. In an experimental study, the proposed method is shown to be feasible in industrial heterogeneous networks and provide better security paths without affecting availability in the multi-domain and multi-nodes case of industrial networks.

Wireless Sensor Networks (WSNs) have rapidly increased to be applicable in many different areas due to their wireless mobile connectivity, large scale deployment and ad hoc network. However, these characteristics make WSNs usually deployed in unattended and hostile field, which may bring some new threats such as information tampering and eavesdropping attacks, etc. User authentication is one of the most important security services that allowed the legitimate user to query and collect the real-time data from a sensor node in WSN. Since the sensor nodes are resource-constrained devices which have limited storage, power and computing resource, the proposed authentication scheme must be low cost and lightweight. Recently, Gope et al. proposed a realistic lightweight authentication for real-time data access in WSN. Unfortunately, through our analysis we identified several flaws exist in their scheme as well as exist in other two-factor schemes. In order to withstand these threats, in this paper, we proposed some solutions to withstand the problems in Gope et al.’s and other schemes. Our solutions provide an important reference for security data access in WSN.

With the deepening integration of informatization and industrialization, the industrial control system is facing more and more serious security threats at the same time of rapid development. At present, the legal norms and national security standards in the field of industrial control system are relatively lacking. And there is no strict market access system. In addition, the state’s industrial support for domestic industrial control equipment needs to be strengthened. Especially in terms of system security, data security, application security, and security management system, the research investment needs to be further increased, professional and technical forces need to be cultivated and the research of core technologies need to be focused on. To solve the information security problem of industrial control systems has become one of the key topics that the industry pays close attention to. In this paper, the development history of industrial control system is introduced, the root cause of industrial control system security threats is deeply analyzed, the future security threats of industrial control system is pointed out and the safety precautions of industrial control system is put forward. Based on the analysis of this paper, the personal security awareness of the industrial control system can be raised and the challenges of security threats can be better solved.

Blockchain, as the core technology of cryptocurrency, provides a novel idea for the sharing economy. A user pays the agreed money to the property owner without any trust third party in blockchain, which leads to the leakage of privacy due to openness of the blockchain. To alleviate the privacy concern, a protocol is proposed to not only protect privacy through breaking the link between the user and the property owner but also ensure the fairness among parties joined in transaction. In addition, double-spending and double-usage are detected and prevented by applying smart contract in the protocol. Compared with the existing related works, the proposed protocol is secure, effective and practical.

Smart transportation system is a cross-field research topic involving a variety of disciplines, in which data plays a central role. Researches that are driven by data can be traced back to the 1930s, when the British statistician and biologist Ronald Fisher creates the Iris dataset to study the objective and automated way to classify iris flower. Early success of data powered research illustrates the potential value of data in the research topics in either scientific or social domains. City transportation system is one of the most fundamental components of the city service. Recent researches show that the quality of the transportation service largely depends on how well its resources can be managed and utilized, which in turn relies on how well the data derived from that system can be collected and processed for the need of the government authority, as well as any individual citizen. Improvements on the transportation via the smart transportation system do not only pose an important impact on any individual’s life style, but it is also a great saving of time and energy.

Internet of Things, or IoTs have gained popularity in the recent days for interconnecting the things such as devices, sensors, equipment, software, and information services. Besides, IoT has played a remarkable role in all aspects of daily lives, which covers many fields including healthcare, automobiles, entertainments, industrial appliances, sports, and homes. IOT has also much effect on smart fields, such as smart homes, smart cars, and so on. These smart technologies are not only used at homes but also used in various other sectors such as media, business, agriculture, securities, and transportation. But compared with traditional networks, the sensitive nodes of the IoT are assigned in positions without manual supervision, with the weak capability and limited resources, making the security issues of the IoT quite troublesome, and in addition, the fast development and wider adoption of IoT devices signify the urgency of addressing these security threats before deployment. This paper provides the readers with a basic understanding of IoT, such as its origin, classification, underlying technologies, various applications, and security issues and challenges that IoT is facing with, it also enables the application developers and security researchers to acquire the current development status about the IoT.

In recent years, the developments in electronic science and technology have led to abundant new wearable devices and IMDs, but with the use of those devices increases, the concern about their security also becomes serious. In this paper, we give a survey on the security threat and protection method of IMD and wearable devices and compare those protection methods regarding different factors.

This paper introduces the design of key node information collection system for the North-to-South Bulk Grain Container Transportation Project in China. Based on the idea of the Internet of Things (IoT), we combined the current sensing and communication technologies to develop a vehicle-borne information acquisition system and a key node handheld information acquisition system. This project researches on smart front-end products, low-power sensor technology, data acquisition, and wireless communication technology. The whole system combines cloud platform, edge computing, and other technologies to fully utilize the Internet of things resources and artificial intelligence. It demonstrates and certifies the product performance of key node information collection equipment for bulk grain containers. The system adapts to the public railroad and water transport mode while the entire process of food transportation is monitored through intelligent sensor, GPS, LoRa and 3G/4G technologies so that to achieve information traceability in the grain transportation, thus ensuring food quality.

Ocean exploration is playing an increasingly significant role in the development of each country because of the huge material resources therein. A terrain-aided strapdown inertial navigation system based on Kalman Filter (KF) is proposed in this paper in order to improve the navigation precision of autonomous underwater vehicles. The characteristics of strapdown inertial integrated navigation system and terrain-aided navigation system are described, and improved ICCP method is applied to the terrain aided navigation system. Simulation experiments of novel integrated navigation system proposed in the paper were carried out comparing to the traditional ICCP method. The simulation experiments suggest that the improved method is able to improve the long-time navigation precision relative to the traditional method.

Aiming at the relatively low level of solar greenhouse intelligent control in northern China, while considering the data security of the IoT, this paper describes the construction of solar greenhouse control system from four aspects, including hardware design, lower computer program, upper computer program and mobile APP. The control system has been applied in the greenhouse of Lingxian, Dezhou. The use of the system saves the labor force, improves the labor efficiency and farmer’s income.

The cable is a very important tool for information transmission, and the process of laying by ships will be affected by the waves and wind on the sea, in the paper we set up the model of the process, We now by setting the united equipment acquisition cable on each node in the sea water pressure value, according to the motion state on the cable at the same time establish the physical dynamic equation by finite element method (fem) of matlab simulation to determine the movement of the cable to get sea cable in different sea conditions in case of trajectory can provide a scientific reference for laying optical cable and establish the corresponding equations of motion, and simulate the process of laying under different sea conditions in MATLAB.

In response to various problems existing in the current duck product industry chain. For example, decentralized of the industry chain, lack of data exchange between each link, lack of unified information collection equipment, and so on. An intelligent information management platform for duck product chain was developed, based on the specific needs of each link of the duck and poultry industrial chain. The platform is mainly composed of the information collection equipment and the intelligent management systems. The information acquisition equipment comprehensively uses the current well-perceived and reliable transmission technology of the Internet of Things to achieve the seamless collection of information in each link of the duck product industry chain and the seamless convergence of information in each link. The intelligent management system utilizes big data analysis technology to realize internal automation and digital management. The long-term test of the system shows that the data in each link of the system seamlessly connects. The data collected by the system is accurate and reliable. Its operation is simple and convenient. The system is highly scalable and suitable for use in production.

Traditionally, it is considered that there is no sensitive information in the management frames and only the data frames need encryption protection in the initial 802.11 standard protocol, so no corresponding security mechanism is required in the management frames. But with the popularity of WLAN, the researchers realize that the lack of the management frames security mechanism can lead to many security problems. For example, an attacker can constantly transmit fake management frames to dissociate normal connection between the AP and the legitimate station. Therefore, the management frames security mechanism must be established.In this paper, we propose a topic model based method to realize the authentication of Wi-Fi management frames using the Channel State Information (CSI). CSI values have a strong correlation with location. That means CSI values between the AP and the legitimate station have significant distinction with the CSI values between the AP and the attacker. We utilize the topic model – an unsupervised machine learning method to extract features automatically from the collected CSI values. Extensive experiments are conducted to analysis the system performance. The experiments prove that topic model can better resist noise interference than the traditional method and achieve average 91% accuracy of the management frame authentication.

The phenomenon of inconsistent network properties such as forwarding loops and packet losses may occur during rule update of software-defined wireless sensor networks. Aiming at this, the idea of scheduling the update order of sensor nodes is adopted to design the rule reverse addition strategy, rule suppression modification strategy and rule obverse deletion strategy. At the same time, parallel execution of sub-updates is supported by dividing the update domain. This method effectively improves the consistency of network update attributes and greatly enhances the update performance. The simulation results based on SDN-WISE show that the proposed strategies could adapt well to the resource-limited wireless sensor networks, and the advantages are especially obvious when the rule dependencies are more complicated.

Multi-party applications are becoming popular due to the development of mobile smart devices. In this work, we explore PVKE (Physical Vibration Keyless Entry), a novel pairing mechanism, through which users are able to utilize smart devices to detect a vibration from the smart watch, so as to use information from the vibration to deconstruct security keys. Thus, we perform device pairing without complicated operations. We show that the recognition accuracy of vibration key detection achieves 95% through solid experiments.

Where IoT and Cloud Computing are revolutionizing today’s ecosystem, they also cause alarming security and privacy issues. As a continuum of devices and objects are interconnected with each other in order to share data and information. It is really important to evaluate the trustworthiness of these devices/objects, whether they are trustworthy or malicious. In this work, we propose a novel Trust and Reputation (TR) based model to outsource malicious users in a Fog-based IoT (FIoT). In our model, we used a multi-source trust evaluation by taking into account of the reputation of participating nodes. We use the feedback system to make the trust system reliable and trustworthy. We evaluate our model with simulations and the result shows the effectiveness of TRFIoT.

This paper investigates the location security problem of WLAN location systems under multipath communication scenarios. As the location information is critical in some location related applications, it is important to guarantee the user’s location security. The scheme that based on the signal’s arrival angle is an important kind of the location strategies. The selection of the direct path is critical to the accuracy of the system. We analyse the vulnerability of the angle based system. We show that if the system selects the incorrect direct path, it will potentially lead to a false position. By detailed simulation which based on quasi-mirror reflection model, we demonstrate that the manipulated false direct path can lead to error position effectively.

The main body of this article shows a system designed to improve the level of greenhouse irrigation and fertilization automation in China, while improving the utilization of water and fertilizer resources. Tomato greenhouse irrigation was used as an example to design an intelligent control system for tomato water and fertilizer integration in the greenhouse. The system is based on the optimal growth conditions of greenhouse tomatoes, combined with the impact of time on greenhouse tomatoes, using NB-IOT technology, information processing technology, information acquisition technology and other networking technology. The system has two sub-modules: tomato water and fertilizer integration intelligent control module and intelligent warning module. The system improves the precision of smart irrigation and fertilization in greenhouse tomatoes. In addition, in order to facilitate the user’s real-time supervision of the greenhouse, the system is equipped with a dedicated mobile phone APP. Experiments have shown that the system greenhouse environment information collection and uploading are very stable, and tomato growth environment is controlled reliably. The system satisfies the needs of modern greenhouse planting management, and significantly improves the benefits of tomato production.

3D objects are becoming ubiquitous while being used by many mobile and social network applications. Meanwhile, such objects are also becoming a channel being used for covert communication. Steganalysis aims to identify when information is transferred in such ways. This research study analyses the influence of the 3D object smoothing, which is an essential step before extracting the features used for 3D steganalysis. During the experimental results, the efficiency when employing various degrees of 3D smoothing, is assessed in the context of steganalysis.

Segmenting brain magnetic resonance (MR) images accurately is of great significance to quantitatively analyze brain images. However, many traditional segmentation methods underperforms due to some artifacts in brain imaging such as noise, weak edges and intensity inhomogeneity (also known as bias field). Recent methods based on convolutional neural networks (CNN) suffer from the limited segmentation accuracy for details. To settle these problems and to obtain more accurate results, a modified U-Net model is proposed in this paper. Different sized filters are used in every single conv-layer. The outputs of different filters are concatenated together and then the concatenated feature maps are fed to the next layer. This work makes the network learn features from different scales, be-sides, it also reduce the filter space dimensionality. Both experimental results and statistic results show that our model have higher accuracy and robustness in segmenting brain MR images.

Cyber-Physical-Social System (CPSS) involves numerous connected smart things with different technologies and communication standards. While CPSS opens new opportunities in various fields, it introduces new challenges in the field of security. In this paper, we propose a real-time forensics-aware scheme for supporting reliable image forensics investigations in the CPSS environment. The forensic scheme utilizes a multichannel convolutional neural network (MCNN) to automatically learn hierarchical representations from the input images. Most previous works aim at detecting a certain manipulation, which may usually lead to misleading results if irrelevant features and/or classifiers are used. To overcome this limitation, we extract the periodicity property and filtering residual feature from the image blocks. The multichannel feature map is generated by combining the periodic spectrum and the residual map. Micro neural networks module is utilized to abstract the data within the multichannel feature map. The overall framework is capable of detecting different types of image manipulations, including copy-move, removal, splicing and smoothing. Experimental results on several public datasets show that the proposed CNN based scheme outperforms existing state-of-the-art schemes.

Multiset integrated canonical correlation analysis (MICCA) can distinctly express the integral correlation among multi-group feature. Thus, MICCA is very powerful for multiple feature extraction. However, it is difficult to capture nonlinear relationships with the linear mapping. In order to overcome this problem, we, in this paper, propose a multi-kernel multiset integrated canonical correlation analysis (MK-MICCA) framework for subspace learning. In the MK-MICCA framework, the input data of each feature are mapped into multiple higher dimensional feature spaces by implicitly nonlinear mappings determined by different kernels. This enables MK-MICCA to uncover a variety of different geometrical structures of the original data in the feature spaces. Extensive experimental results on multiple feature database and ORL database show that MK-MICCA is very effective and obviously outperforms the single-kernel-based MICCA.

With the development of digital image processing and big data analysis, there exist massive digital images that can be obtained by ordinary online users. However, regarding to the volume of data and the lack of access control embedded within digital images, some images may impose risks of leakage and exposure. Especially, some images need to keep secret totally or at least only can be accessed partially. Those are not be classified and protected with a subtle manner or accessed by fine-grained privileges. In addition, recent most access control strategies separate control lists from accessed objects, so that the execution of access control relies on the networking links to remote servers. Sometimes remote servers may be available, thus the control will be hindered. Moreover, it is not easy to share and distribute a large volume of data together with their access control policies on different servers. In this paper, we propose a novel access control model based on invisible watermarks, where embed access control policies. This model realizes the marriage of accessing objects and access control strategies, and can be delivered together with big data objects such as digital images.

Typically, people would visit travel websites such as tripadvisor.com , mafengwo.cn , or ctrip.com when planning for their next trip. The lowest airfare, the best hotels, and great attractions can be found on these websites based on requirements provided by users. Millions of traveler reviews, photos, and maps, are also available. With all this information, it may still be time-consuming for users to work out a travel plan, which involves section of attractions from a huge candidate list, and more importantly, an itinerary that guides their daily activities. We therefore proposed a recommendation technique that facilitates the creation of personalized travel plans. Using a tag-based mapping algorithm, we create a list of candidate attractions that best match with the user favorite spots. An itinerary containing attractions that are most appealing to users will be derived from the candidate list and we refer to this kind of itinerary as MAI. Meanwhile, by applying K-Means clustering to the list of candidate attractions according to their geographical location, we will be able to produce the shortest itinerary (SI) and the itinerary with the highest performance/price ratio (MEI). A series of experiments have been carried out to help evaluation of our recommendation technique and the results demonstrate that our personalized recommender for travel planning can provide a better and more detailed travel plan that satisfies users with various requirements.

The popularity and portability of high-fidelity recording devices and playback devices pose severe challenges for speaker recognition systems against replay voice attacks. In this paper, the signal of audio is transformed into the frequency domain through the Fourier trans-form and constant Q transform. Compared with genuine voice, the mean and standard deviation of the replay voice at each frequency bin has changed slightly. And through the coefficient of variation to further analyze the difference between genuine voice and replay voice. A detection algorithm based on fusion feature is proposed. The algorithm uses two kinds of time-frequency transform coefficients and their cepstrum characteristics to train the GMM model and calculate the likelihood ratio score. Finally, the replay voice is detected by the fusion of scores. The experimental results show that the algorithm is about 13% lower than the baseline EER provided by The ASV Spoof 2017.

Recoverable speech forensics algorithm not only can locate the attacked frames, but can reconstruct the attacked signals. Meanwhile, the method can provide useful information for the prediction of attacker and attacker’s intent. We proposed a robust recoverable algorithm used for digital speech forensics in this paper. We analyze and conclude that large amplitude DCT coefficients play a more significant role for speech reconstruction. Inspired by this, we regard the large amplitude coefficients as compressed signal, used for the reconstruction of attacked frames. For embedding, we scramble samples of each frame, and embed frame number and compressed signal into less amplitude DCT coefficients of scrambled signal by substitution. Frame number is used for tamper location of watermarked speech, and compressed signal is used for the reconstruction of attacked signals. Experimental results demonstrate that the algorithm is inaudible and robustness to signal processing operations, has ability of tamper recovery and improves the security of watermarking system.

With compressed speech codecs having been widely used as the key technology for VoIP and mobile communications, steganography based on compressed speech streams has been flourishing in recent years. In this paper, based on the codec’s characteristic features, we present a steganographic method using the excitation pulse positions of the high bit rate speech codec of ITU-T G.723.1. In the proposed method, the codes of excitation pulse-positions in the speech codec (i.e., codebook positions) are finely modulated according to the secret information that is being hidden. Sensitive data may thus be transmitted in secret without affecting the transmission quality of the normal speech data. Using the algorithm developed in this work and with the all-odd/even characteristics of pulse code positions being utilized, steganography experiments at high bit rates (6.3 Kbit/s) ware conducted on four kinds of voices: male or female voices speaking Mandarin or English. By using the proposed approach, an embedding rate of 2.6% and secret information transfer rate of 166 bits/s resulted in <5.0% degradations of the Perceptual Evaluation of Speech Quality (PESQ) score. And when the data hiding capacity reached 8.8% and the rate of secret information transfer came to 566 bits/s, the PESQ score was still reduced by <12.1%. The experiments show that our algorithm performs a higher degree of secrecy and steganographic efficacy compared with existing similar algorithms.

The computational simulations under the two dimensional hydrophobic-polar (2D-HP) model from protein’s amino is a fundamental and challenging problems in computational biology. In this paper, we propose an improved quantum-inspired evolutionary algorithm based on variable angle-distance rotation strategy (QEA-VAR) for this NP-hard combinatorial protein folding problems. The QEA-VAR method is based on the concept and principles of quantum computing, such as quantum bits, quantum rotation gates and superposition of states. Comparing to the previously well-known evolutionary algorithm for the protein folding problem, QEA-VAR can find optimal or near-optimal energy structure from the benchmark sequences with a small simulating samples. Moreover, the proposed method’s global convergence is faster than the other evolutionary algorithms. The application studies have demonstrated the superior performance and feasibility of the proposed algorithm for the protein folding problem.

To solve the problems that there is no sufficient annotated data in low-resource languages and it is hard to mine the deep semantic correspondence between languages via existing bilingual word embedding learning methods, this paper presents an effective text processing method based on transfer learning and bilingual word embedding model CWDR-BiGRU (Cross-context window of dynamic ratio bidirectional Gated Recurrent Unit) which contains an enhanced Skip-gram called cross-context window of dynamic ratio and encoder-decoder. The method can process low-resource language text effectively only using sentence-aligned corpus of bilingual resource languages and annotated data of high-resource language. The experimental results of semantic reasoning and word embedding visualization show that CWDR-BiGRU can effectively train bilingual word embeddings. In the task of Chinese-Tibetan cross-lingual document classification, the accuracy of transfer learning method based on CWDR-BiGRU is higher than the conventional method by 13.5%, and higher than the existing Bilingual Autoencoder, BilBOWA, BiCCV and BiSkip by 7.4%, 5.8%, 3.1% and 1.6% respectively, indicating CWDR-BiGRU which has reduced the difficulty of acquiring corpora for bilingual word embeddings can accurately excavate the deep alignment relationship and semantic properties.

Audio compression history detection is a significant part of audio forensics, which is important to detect whether audio has been tampered or forged. When the structure of AAC audio frame is destroyed, its spectral coefficient distribution is similar to that of audio after the first compression. An algorithm of AAC audio compression detection was presented by using the statistical characteristics of QMDCT coefficients before and after removal of sampling points as the discriminative feature. Experimental results demonstrate that the proposed method can distinguish the single, double compressed AAC audios effectively, and from the low-bit-rate transcoding to high-bit-rate, the average classification accuracy achieves 99.84%, the same-bit-rate compression detection accuracy achieves 98.60%. In addition, the results of comparison experiments show that our algorithm outperforms the state-of-the-art algorithm.

For the problem of low manual detection accuracy under the conditions of illumination and occlusion, the detection of human hands based on common optical images was explored, and an accurate manual detection method under general conditions was proposed. The method based on skin color model combined with Convolutional Neural Network (CNN) was mainly used. Realize the detection of human hands. Firstly, the skin color model is obtained according to the characteristics of skin color in the HSV (Hue, Saturation and Value) space, which is used to segment skin area. On this basis, a convolutional neural network for the detection of human hand contours is constructed, which is used to extract the human hand contour features to constrain skin region to obtain the hand region. The results show that even in light and shielding, it also has adaptability, which improves the accuracy of hand detection.

The process of digital image acquisition and transmission is easy to be polluted by noise. Noises can also cause disturbances, or even misjudgements in the remote sensing image, face recognition, image classification of machine learning and deep learning. Therefore the correctness and safety of image usage is greatly reduced. Different types of noise may occur under various conditions, and the same filtering method has different effects on different types of noise processing, which makes it difficult to select the best way to filtering the image. So the detection and recognition of noise type has always been a hot topic in the field of information security. However, there are lacking solutions to the current noise type identification problem, and the complexity is very high. In this paper, a convolutional neural network(CNN) model which is able to automatically identify salt and pepper noise, Gauss noise and random noise based on deep learning training is proposed. After that, median filter, mean filter and wiener filter are used to filter the corresponding images. The purpose of ensuring the correctness and security of the image application is achieved. By simulating the images of different noise and analyzing PSNR, it is proved that this method able to distinguish the noise and filter obviously.

The Audiovisual Speech Recognition (AVSR) most commonly applied to multimodal learning employs both the video and audio information to do Robust Automatic Speech Recognition. Traditionally, AVSR was regarded as the inference and projection, a lot of restrictions on the ability of it. With the in-depth study, DNN becomes an important part of the toolkit in traditional classification tools, such as automatic speech recognition, image classification, natural language processing. AVSR often use some DNN models including Multimodal Deep Autoencoders (MDAEs), Multimodal Deep Belief Network (MDBN) and Multimodal Deep Boltzmann Machine (MDBM), which are always better than the traditional methods. However, such DNN models have several shortcomings: Firstly, they can’t balance the modal fusion and temporal fusion, or even haven’t temporal fusion; Secondly, the architecture of these models isn’t end-to-end. In addition, the training and testing are cumbersome. We designed a DNN model—Aggregate $$\varvec{d}$$ Mult $$\varvec{i}$$ moda $$\varvec{l}$$ Bidirection $$\varvec{a}$$ l Recurren $$\varvec{t}$$ Mod $$\varvec{e}$$ l (DILATE)—to overcome such weakness. The DILATE could be not just trained and tested simultaneously, but alternatively easy to train and prevent overfitting automatically. The experiments show that DILATE is superior to traditional methods and other DNN models in some benchmark datasets.

The linearly separable problem is a fundamental problem in pattern classification. Firstly, from the perspective of spatial distribution, this paper focuses on the linear separability of a region dataset at the distribution level instead of the linearly separable issue between two datasets at the traditional category level. Firstly, the former can reflect the spatial distribution of real data, which is more helpful to its application in pattern classification. Secondly, based on spatial geometric theory, an adaptive construction method for testing the linear separability of a region dataset is demonstrated and designed. Finally, the corresponding computer algorithm is designed, and some simulation verification experiments are carried out based on some manual datasets and benchmark datasets. Experimental results show the correctness and effectiveness of the proposed method.

Resolution coefficient of traditional gray relation classifier usually takes a fixed value of 0.5, which greatly limits the adaptive ability, and reduces the effectiveness of this algorithm to identify signals. To solve this problem, an improved optimized resolution coefficient algorithm of gray relation classifier was proposed. Particle swarm optimization (PSO) algorithm was used to calculate the optimized resolution coefficient corresponding to the best classification results under different SNR environment. The adaptive ability of this algorithm was improved by improving the selection method of resolution coefficient and ultimately the classification effect was improved. Simulation results show that, compared with the traditional improved algorithm, it can improve the recognition rate of signals under different SNR environment, and have a good application value.

The formation of the traffic accidents is always caused by many factors such as the road, the vehicle, the drivers conditions and so on [1, 2]. In order to analyze the various causes of traffic accidents, the feasibility measures should be taken after the accident occurred, and how to prevent the occurrence of related traffic accidents. This paper introduces a novel traffic accident analysis system to analyze traffic accidents, which is mainly composed of seven parts, including accident basic information analysis, accident driver analysis, accident vehicle analysis, accident road analysis, large accident ledger, multi-dimensional accident analysis, and accident analysis report. The suggested framework mainly contains two technology: (1) multi-dimensional analysis which is the core of the data warehouse technology to build the multi-dimensional data model in reservoir management (2) the on-line analytical processing (OLAP) technology in data analysis and display. In addition, Bayesian network is used for multidimensional data analysis in this paper. The method proposed in this paper can effectively organize a large number of out of data to get useful information which is convenient for traffic management departments to analyze the reason of traffic accidents and then to take corresponding countermeasures.

Currently, the hand tracking based on deep learning has made good progress, but these literatures have less influence on the tracking accuracy of Convolutional Neural Network (CNN) architecture and parameters. In this paper, we proposed a new method to analyze the influence factors of gesture tracking. Firstly, we establish the gesture image and corresponding gesture parameter database based on virtual 3D human hand, on which the convolutional neural network models are constructed, after that we research some related factors, such as network structure, iteration times, data augmentation and Dropout, etc., that affect the performance of hand tracking. Finally we evaluate the objective parameters of the virtual hand, and make the subjective evaluation of the real hand extracted in the video. The results show that, on the premise of the fixed training amount of the hand, the effect of increasing the number of convolutional cores or convolution layers on the accuracy of the real gesture is not obvious, the data augmentation is obvious. For the real gesture, when the number of iterations and the Dropout ratio is about 20%–30%, good results can be obtained. This work provides the foundation for future application research on hand tracking.

Describing videos in human language is of vital importance in many applications, such as managing massive videos on line and providing descriptive video service (DVS) for blind people. In order to further promote existing video description frameworks, this paper presents an end-to-end deep learning model incorporating Convolutional Neural Networks (CNNs) and Bidirectional Recurrent Neural Networks (BiRNNs) based on a multimodal attention mechanism. Firstly, the model produces richer video representations, including image feature, motion feature and audio feature, than other similar researches. Secondly, BiRNNs model encodes these features in both forward and backward directions. Finally, an attention-based decoder translates sequential outputs of encoder to sequential words. The model is evaluated on Microsoft Research Video Description Corpus (MSVD) dataset. The results demonstrate the necessity of combining BiRNNs with a multimodal attention mechanism and the superiority of this model over other state-of-the-art methods conducted on this dataset.

The hydropower dam construction and transformation of research involves the social politics, economy, environment and other aspects of content, both in theoretical system and production practice is of great significance in this paper, from the perspective of the cascade reservoirs, the purpose is to use the method of probability theory, analyzing the various factors influencing the design flood of cascade reservoirs, and transformation of cascade reservoirs are studied by using JC method determine the principle of design flood, and using the theory of the zambezi river Carrie and the replacement of the dam for the design of the simulation and experiment.

Although matrix factorization has been proven to be an effective recommendation method, its accuracy is affected by the sparsity of the matrix and it cannot resolve the cold start problem. Social recommendation methods have attracted much attention in solving these problems. In this paper, we focus on community discovery rather than individuals’ relations in the social network and propose a community-based matrix factorization (CommMF) model. It consists of two parts. One is a community detection algorithm Coo-game, proposed in our previous work and used here to divide the social network of users into multiple overlapping communities. It is based on the game theory and can fast detect overlapping communities. Since the users in the same community share the common interests such as scoring information, some of the null values in the scoring matrix can be filled according to the communities. This will help alleviate the sparsity of the scoring matrix and the cold start problem of new users. The other part is the matrix factorization model, which is used to recommend items to users. The model is trained by a stochastic gradient descent algorithm. The experimental results on real and simulated datasets show that CommMF can get higher accuracy with the help of community information compared with PMF and SocialMF models.

Local descriptors are very effective features in bag-of-visual-words (BoW) and vector of locally aggregated descriptors (VALD) models for image retrieval. Different kinds of local descriptors represent different visual content. We recognize that spatial contextual information play an important role in image matching, image retrieval and image recognition. Therefore, to explore efficient features, firstly, a new local composite descriptor is proposed, which combines the advantages of SURF and color name (CN) information. Then, VLAD method is used to encode the proposed composite descriptors to a vector. Third, local deep features are extracted and fused with the encoded vector in the image block. Finally, to implement efficient retrieval system, a novel image retrieval framework is organized a novel image retrieval framework is organized based on the proposed feature fusion strategies. The proposed methods areis verified on three benchmark datasets, i.e., Holidays, Oxford5k and Ukbench. Experimental results show that our methods achieves good performance. Eespecially, the mAP and N-S score achieve 0.8281 and 3.5498 on Holidays and Ukbench datasets, respectively.

Detecting surface defects of alumina substrate by using computer technique will enhance productivity in industrial manufacture. Edge detection of image is the commonly used technique for the detection of surface defects. However, it is difficult to automatically detect the surface defects of the alumina substrate since the noise and the multiple kinds of defects may exist in a substrate. In this paper, we designed an edge detection algorithm based on Canny detector aiming to automatically detect the surface defects of alumina substrate. Our algorithm can adaptively smooth image as well as adaptively determine the low threshold and high threshold. Experiments show that our algorithm can effectively and automatically detect several kinds of surface defects in the alumina substrate.

In fact, microblog users’ influencing ability is the same thing to their spreadability. The research of microblog users’ spreadability has received much attention in the past years, towards which most of the existing advanced methods are based on the spread range of users’ blogs. However, those methods neglect the blogs’ disseminating quality, we find that only the spread range of those blogs with disseminating quality could precisely indicate the spreadability of microblog users, yet those without will jeopardise the precision of the analysis of users’ spreadability. In addition, the information dissemination in microblog mainly relies on the followers’ reposting behavior, yet the existing methods have not taken much into consideration of the factors that affect the followers’ reposting behavior. In this paper, the blogs’ disseminating quality in microblog is thoroughly studied. In addition, we propose a method to measure microblog users’ spreadability based on the forecast of the spread range of users’ blogs possessing disseminating quality. In the analysis, we innovatively combine the users’ attention degree received from their followers and interest similarity between them. We conducted several experiments on the real data set of Sina Weibo, a very common and popular microblog in China, and the results prove the accuracy of our method.

Textual emotion analysis is an important research issue in natural language processing. In this paper, we address a novel task on emotion, called emotion effect detection, which aims to identify the effect event of a particular emotion happening. To tackle this task, we propose a two-stage model which consists of two components: the identification module and the extraction module. In detail, the identification module aims to judge whether a sentence group contains emotion effect, and the extraction module aims to extract the emotion effect from a sentence group. These two modules are learned with maximum entropy and conditional random field (CRF) methods respectively. Empirical studies demonstrate that the proposed two-stage model yields a better result than the one-stage model.

Assessing the influence relevance of a news article is a very important and novel task for news personalized recommendation services. It provides a novel functionality by additionally recommending users news articles that may not match users’ interest points but can help users make good decisions in their daily lives. Since the influence of implicit information delivered by news articles cannot be obtained literally, and meanwhile regions and industries affected by the influence of implicit information are usually not explicitly mentioned in news articles, machine-based methods lost their ability. In this paper we explore methods of assessing influence relevance of news articles by employing crowdsourcing, and the experimental results show that crowdsourcing can assess the influence relevance of news articles very well.

Traditional steganography algorithms embed secret information by modifying the content of the images, which makes it difficult to fundamentally resist the detection of statistically based steganalysis algorithms. To solve this problem, we propose a novel generative steganography method based on generative adversarial networks. First, we represent the class labels of generative adversarial networks in binary code. Second, we encode the secret information into binary code. Then, we replace the labels with the secret information as the driver to generate the encrypted image for transmission. Finally, we use the auxiliary classifier to extract the label of the encrypted image and obtain the secret information through decoding. Experimental results and analysis show that our method ensures good performance in terms of steganographic capacity, anti-steganalysis and security.

The influence of consumer reviews on hotel reservation has been addressed in both practical and theoretical fields. Most previous studies focus on the volume and score of reviews, little about the score of hotel attributes. This paper focuses on that how the variance of the hotel dominance attribute ratings affects the consumer recommendation rate to economy hotels and luxury hotels respectively. Research model has been developed based on Elaboration Likelihood Model and tested with the data from Ctrip.com in China. The results show that the variance of economy hotels has positive effect on the consumer recommendation rate, while that of luxury hotels has negative effect. Furthermore, results also indicate that the volume and score of reviews have different moderating effect.

Low-rank prior knowledge has indicated great superiority in the field of image processing. However, how to solve the NP-hard problem containing rank norm is crucial to the recovery results. In this paper, truncated weighted schatten-p norm, which is employed to approximate the rank function by taking advantages of both weighted nuclear norm and truncated schatten-p norm, has been proposed toward better exploiting low-rank property in image CS recovery. At last, we have developed an efficient iterative scheme based on alternating direction method of multipliers to accurately solve the nonconvex optimization model. Experimental results demonstrate that our proposed algorithm is exceeding the existing state-of-the-art methods, both visually and quantitatively.

Semantic annotation has attracted a growing interest in the information retrieval and computer vision. Existing methods have typically focused on several visual cues and semantic context information with an image itself using different frameworks, neglecting the prior knowledge constraints about the real world. However, strong prior knowledge embedding should be considered to improve the performance of semantic annotation tasks. Note that semantic objects will interact each other during the semantic prediction stage, and the support visual relationships can affect the recall and accuracy of semantic annotations. In this paper, we exploit a novel method to semantic modeling with prior knowledge embedding to jointly find the semantic objects and the corresponding support relationships in the images. Inference in the model can be conducted exactly via graph modeling and knowledge embedding, and the parameters can be learned at the supervised learning stage. The extensive experiments on COCO15 and Stanford Visual Relationship data sets confirm the benefits of semantic annotation for the objects for the knowledge embedding.

Malicious web pages that use drive-by-download attacks or social engineering technique have become a popular means for compromising hosts on the Internet. To search for malicious web pages, researchers have developed a number of systems that analyze web pages for the presence of malicious code. Most of these systems use dynamic analysis. That is, the tools are quite precise, the analysis process is costly. Therefore, performing this analysis on a large-scale of web pages can be prohibitive. In this paper, we present JSPRE, an approach to search the web more efficiently for pages that are likely malicious. JSPRE proposes a malicious page collection algorithm based on guided crawling, which starts from an initial URLs of know malicious web pages. In the meanwhile, JSPRE uses static analysis techniques to quickly examine a web page for malicious content. We have implemented our approach, and we evaluated it on a large-scale dataset. The results show that JSPRE is able to identify malicious web pages more efficiently when compared to crawler-based approaches.

The convolution neural networks (CNNs) can extract the rich feature of the image. It was widely used in the field of computer vision (CV) and made great breakthroughs. However, most of the existing CNNs models only utilize the features out put by last layer, the representation of features is not comprehensive enough. In this paper, we propose a multilevel features fusion method, in order to make full use of the intermediate layer features. This method can strengthen feature propagation and improve the accuracy of downstream tasks. We evaluate our method through experiments on two image classification benchmark tasks: CIFAR-10 and CIFAR-100. The experimental results show that our method is able to significantly improve the accuracy of VGG-like model. The improved model is better than most existing models.

The assessment of individual personality traits plays a crucial role in important social events such as interpersonal relationships, job search, crime fighters, and disease treatment. In this paper, a multi-view (frontal and profile view, 2.5D) facial feature extraction model is proposed to evaluate the possible correlation between personality traits and face images. Our main contribution and innovation are threefold: Our primary contribution is the development of a 2.5D hybrid personality computational model in order to gain a more comprehensive understanding of one’s personality traits; the second is that we have established two datasets. Datasets of people over 35 years of age are compared with those of 16–35-year-olds. We focus on the relationship between personality traits in human face and age; Finally, on a dataset of 500 facial images pre-processed from our face database and an personality score dataset collected from human testers, we evaluate the model through the application of support vector regression (SVR). Result shows that the prediction performance of the 2.5D feature model is better than that of the 2D model. We show that the 2.5D model performs well with low statistic error (MSE = 0.4991) and good predictability (R2 = 0.5638).

With the rapid development of image display technology and digital acquisition device, recapturing the copies of images from LCD screens with high quality becomes rather convenient and easy. Such recaptured images post severe security threats in image credibility and bio-authentication. In order to alleviate such security problems, in this paper, we propose a simple yet effective method to detect the images recaptured from LCD screens based on enhanced residual-based correlation coefficients. Specifically, we first extract the blocks which contain only one edge with pre-defined criteria. Then, pixel-wise sharpness is used to enhance the discriminability of the extracted single-edge blocks. Finally, pixel-wise correlation coefficients in the residual of the enhanced single-edge blocks are adopted as features. Extensive experiments on three high-resolution and high-quality recaptured image databases demonstrate the superior of our proposed method when compared with the state-of-the-art approaches.

Reference-sharing based self-embedding watermarking schemes had been shown to be an effective way to avoid the tampering coincidence and the reference waste problems. Typical reference-sharing based schemes adopt pseudo-random binary matrices as the encoding matrices to generate the reference information. This paper investigate to reduce the computational complexity of the reference-sharing based self-embedding watermarking approach by using the sparse binary matrices as the encoding matrices. Experimental results demonstrate the proposed approach can reduce the computational complexity significantly while maintaining the same tampering restoration capability as the traditional.

The study on the cascading failure of complex network is an important branch in the complex network researches. In this paper, the attack strategies used in the researches on the cascading failure of complex network at home and abroad have been summarized. At the same time, the different characteristics of the new attack strategies and the traditional attack strategies have been analyzed. The modeling principles and the methods of the cascading failure model of complex network have been proposed. Furthermore, the research progress of the cascading failure in the complex network is reviewed. The existing problems as well as the future development trends have been pointed out.

The scientific construction of technical specifications for construction of highway asphalt pavement puts forward higher requirements on the performance and grade of construction materials such as asphalt and other materials. The performance of asphalt mixture has an important impact on the quality of pavement engineering. In this paper, we research and analyze different whetstones and different gradations of dense-graded asphalt mixture by applying computer technology and the Marshall test. the results of the Marshall test show that the properties of dense-graded asphalt mixture are within the scope of the index under a reasonable gradation, and dynamic stability and residual stability meet the specifications, and the dynamic performance and road performance of dense-mixed materials meet the requirements by the comprehensive testing.

The good marine ecological environment is the basis for the sustainable development and utilization of marine resources. However, humans have also severely damaged the marine environment while utilizing marine resources. Therefore, image classification for marine pollution is beneficial to the protection and development of the ocean. In recent years, with the rise of convolution neural networks, this algorithm is rarely used in the classification of marine pollutants. This paper will apply the design of 6-layer convolution neural network to image classification of marine pollution (called for short MP-net). Experiments show that Alex net, VGG(11) and MP-net are learning and training in the same data set, and the accuracy rates respectively are 89.17%, 86.25%, and 90.14%. Therefore, in the image classification of marine pollutants using convolution neural networks, the network can adapt to image scenes, automatically learn features, and have good classification results.

Canonical correlation analysis (CCA) is a typical learning paradigm of capturing the correlation components across multi-views of the same data. When countered with such data with ordinal labels, the accuracy performance yielded by traditional CCA is usually not desirable because of ignoring the ordinal relationships among data labels. In order to incorporate the ordinal information into the objective function of CCA, the so-called ordinal discriminative CCA (OR-DisCCA) was presented. Although OR-DisCCA can yield better ordinal regression results, its performance will be deteriorated when the data are corrupted with outliers because the ordered class centers easily tend to be biased by the outliers. To address this issue, in this work we construct robust manifold ordinal discriminative correlation regression (rmODCR) by replacing the traditional ( $$l_2$$ -norm) class centers with $$l_p$$ -norm centers in objective optimization. Finally, we experimentally evaluate the effectiveness of the proposed method.

Aiming at the limitations of traditional fractal box-counting dimension algorithm in the application of subtle feature extraction of radiation source signals, an improved generalized fractal box-counting dimension algorithm is proposed in the paper. Firstly, the signal is preprocessed, and the real and imaginary data of the signal after Hilbert transform is extracted to obtain the instantaneous amplitude of the signal; Then, the improved fractal box-counting dimension of signal instantaneous amplitude is extracted as the first eigenvector; At the same time, the improved fractal box-counting dimension of the signal without Hilbert transform is extracted as the second eigenvector; Finally, the dual improved fractal box-counting dimension eigenvectors form the multi-dimensional eigenvectors to form a new fractal box-counting dimension eigenvector as signal subtle features, for radiation source signal recognition. By establishing a dual improved fractal box-counting dimension feature space, 11 different practical radiation source signals are classified, compared with the traditional box-counting dimension algorithm, and the recognition rate is calculated. The experimental results show that compared with the traditional fractal box-counting dimension algorithm and the single improved fractal box-counting dimension algorithm, the proposed dual improved fractal box-counting dimension algorithm, can better reflect the signal subtle distribution characteristics under the different reconstruction phase space, and has a better recognition effect with good real-time performance.

The Time Pre-discretized model is firstly adopted to extract web news topic, then a model of topic content evolution is adopted based on the analysis of topic clusters, on which basis a quantification method of topic content is proposed. Experiments on the data sets from social web media find a Pearson correlation coefficient (PCC) of 0.74 between the sequence of topic intensity and that of topic content complexity based on the above quantification method, and a more than 71.5% chance of the simultaneous increase/decrease is observed, showing the “increase or decrease together” law of topic intensity evolution based on topic content evolution.

With the prevalence of various social sites and the rapid development of Internet communication, the problem of team formation in social networks has aroused the enthusiasm of many scholars. Previous research concentrates on raising the variants of this problem and most of them rely on designing approximation optimization algorithms, whose disadvantage is lacking extensibility. In this paper, we deal with the team formation problem for finding a cooperative team within a social network to perform a specific task that requires a set of skills and minimizing the communication cost among team members. In the light of its NP-hard nature, Imperialist Competitive Algorithm (ICA), an evolutionary algorithm for optimization inspired by the imperialistic competition, has been utilized for this problem with different ways of measuring the communication cost. We design a discrete version of ICA by introducing genetic operator in our application, imperialist mutation and imperialist crossover with similarity detection are proposed to avoid a local optimum. Comprehensive experiments on a real-world dataset indicate the performance of the ICA algorithm obtains high-quality results with the comparison of some state-of-the-art ones.

Lightning strike simulation ships, ship body inside and outside surface current distribution of transient and frequency domain, the ship to hold various points transient and frequency domain the electric field and magnetic field distribution and ship to warehouse internal cables on the induction of transient impulse voltage and current and its frequency spectrum. Based on transmission line matrix, the indirect effect of lightning striking the ship is analyzed by CST (Computer Simulation Technology) according to GJB1389A and SAE-ARP5414 relative standards. In this paper, two kinds of lightning stroke paths are simulated by the high current injection. The simulation results show while the lightning stroke paths is not same, the transient magnetic field distribution and current distribution on the surface of the ship is different as well. Moreover, the research method can be used for further study on the protection of the lightning indirect effects and related research. The method can effectively simulate the lightning the lightning indirect effect of ships, for ship lightning protection design and relevant experiment provided the basis and method, has the important value of engineering application.

Facial occlusion, such as sunglasses, mask etc., is one important factor that affects the accuracy of face recognition. Unfortunately, faces with occlusion are quite common in the real world. In recent years, sparse coding becomes a hotspot of dealing with face recognition problem under different illuminations. The basic idea of sparse representation-based classification is a general classification scheme in which the training samples of all classes were taken as the dictionary to represent the query face image, and classified it by evaluating which class leads to the minimal reconstruction error of it. However, how to balance the shared part and class-specific part in the learned dictionary is not a trivial task. In this paper we make two contributions: (i) we present a new occlusion detection method by introducing sparse representation-based classification model; (ii) we propose a new sparse model which incorporates the representation-constrained term and the coefficients incoherence term. Experiments on benchmark face databases demonstrate the effectiveness and robustness of our method, which outperforms state-of-the-art methods.