فصلنامه پردازش علائم و داده ها
سال هجدهم شماره 2 (پیاپی 48، تابستان 1400)

Within the complex Internet of Things (IoT) paradigm that things interact with each other as well as with human beings, one approach to provide security for smart network-based applications is to implement trust management systems.  Trust, in general, overlaps with concepts such as security, privacy and reliability. However, the high number of objects in IoT, along with its dynamic nature and existence of malicious entities, make IoT trust management quite challenging. These unique attributes rule out the application of previous best practices in IoT networks. In this paper, in addition to the analysis of direct, indirect and hybrid trust calculation algorithms, we introduce the relevant attacks and their countermeasures. Moreover, we study the methods for trust model evaluation and also, the effect of limited resources on the performance of trust calculation algorithms.  In short, we carry out a comparative survey in which IoT trust-related works are studied from four perspectives: (1) Trust calculation model, (2) Attack resistance, (3) The effect of resource limitation on the model performance, and (4) Trust management evaluation framework. Through this, we find the pros and cons of the existing algorithms and make a measure for IoT trust management systems. We provide some comparative tables to depict the discrepancies of the existing IoT trust models. One main contribution of this paper is to establish some quantitative metrics to evaluate the trust estimation models and reveal their strengths and deficiencies under different conditions.

Classification is a machine learning method used to predict a particular sample’s label with the least error. The present study was conducted using label prediction ability with the help of a classifier to create a new feature. Today, there are several feature-extraction methods like principal component analysis (PCA) and independent component analysis (ICA) that are widely used in different fields; however, they all suffer from the high cost of transferring to another space. The purpose of the proposed method was to create a higher distinction between various classes using the new feature such that to make the data in the classes closer to each other. As a result, more differentiation is created between the data of various classes to increase the efficiency of classifiers. Firstly, the suggested labels for the primary data set were determined using one or more classifiers and added to the primary data set as a new feature. The model was created using a new data set. The new feature for training and testing data sets was provided separately. The tests were performed on 20 standard data sets and the results of the proposed method were compared with those of the two methods described in the related studies. The outputs indicated that the proposed method has significantly improved the classification accuracy. In the second part of the tests, the resolution of the new feature was examined according to two criteria, namely Information Gain and Gini Index, to examine the effectiveness of the proposed method. The results showed that the feature obtained in the proposed method has higher Information Gain and lower Gini Index in most cases, as it has less irregularity. To prevent the increase in data dimensions, the feature with the least Information Gain was replaced with the feature extracted with the most Information Gain. The results of this step showed an increase in efficiency as well.

Direction of Arrival (DOA) estimation of sound sources using phased array-based methods has a lot of importance in various fields, including sonar, robot vision and mechanical defect detection. Adaptive beamforming methods, such as the MVDR (Minimum Variance Distortionless Response) algorithm, have high resolution compared to non-adaptive methods; but this advantage is achieved in return for the computational complexity of these algorithms. This makes it hard to use these algorithms in applications that require real-time sound source DOA estimation. On the other hand, an important feature of the adaptive beamforming methods including MVDR is the high potential of these algorithms for parallelization. The purpose of this paper is the parallel implementation of the MVDR algorithm by employing GPU instead of CPU to increase the execution speed and achieve real-time mode. To achieve this purpose, the CUDA programming model has been used to implement the algorithm on the GPU. In order to investigate the performance of parallel implementation of the MVDR algorithm, two different GPUs, as well as CPUs, have been used. The performance validity of various implementations in this paper was confirmed by real sonar data as well as simulation data. The results show that using an array of 64 sensors, it is possible to estimate the DOA of underwater sound sources in real-time and with high resolution using the MVDR algorithm.

Trust is one of the most important cornerstones in social networkschr discussions. Most of the times the way that users of these networks trust each other are considered identical, while these users can have different approaches and considerations in trusting to others. Meanwhile, users can impress each other and change their trusting patterns to other users. As a result, the mechanism and manner of impressing opinion trust behavior and conditions of behavioral modes changing have a place of importance to be considered. The question is that, how we can consider different behavior of users and their impression in trusting others? In the first step, the main purpose of this paper is to spotlight social networks different user behavior in trusting other users. For this purpose, the three most important behavioral modes in users) trust are considered. In each of these modes behavioral and functional characteristics of users are the basis of calculating trust, which is based on mental beliefs of them. These modes are named as optimistic, moderate and pessimistic trusting modes. In optimistic mode, we suppose that users think positively and consider low level of activities and signs in trusting others. Here, negative interactions have little impact on users mind. In moderate mode, we suppose that users are not as optimistic as mode A and consider all the interactions and signs when they want to trust others. Here, any negative action can destroy the trust of users and has a greater impact on users. Finally, in pessimistic mode, we suppose that users are pessimistic and trust hardly. In this mode, the interactions that happened more recently have more value than those that happened in the past. In the next step, the way that the trust behavior of users spreads is the goal and innovation of this paper. Three different scenarios are considered for the impressing and spreading of nodes behavior, purposely. In each scenario, different states for users and different purposes for diffusion are defined.   Next, it is followed by maximizing of impression and finding more impressive agents in diffusing trust behavior through social networks. For this purpose, it focused on the structure of users social networks, and the most impressive ones are determined through different diffusion scenarios. The findings of this article appear a significant discrepancy in the amount of trust in each of the different behavioral modes, which is more acceptable in the real world. Analyzing test results leads us to the fact that in the presented model, choosing the start node from each community with 48.14 percent in behavior improvement and diffusion speed and the nodes with the highest degree with 37.03 percent in behavior changing has much more reasonable results than usual models. ​

The most important obstacle for increasing the level of coverage in wireless networks is the energy consumption of the sensors, which causes them to run out of energy. This phenomenon is known as a critical issue called the cover hole, and no sensor is covered in that particular area. As a result, real events in those areas wonchr('39')t be recognizable and traceable. This study has introduced a new algorithm based on the Credit Trust Management System in wireless sensor networks to maximize the coverage of the network by intelligently adjusting the sensor nodes. This algorithm dramatically reduces the amount of coverage by using a backup vector machine while consuming less energy on the network. In other words, by collecting the information in different time periods and interactions between the nodes, their satisfaction is evaluated in order to consider respective rewards or penalties.  Evaluations show that the longevity of the network and the size of the sensor radius of the nodes have improved 12.42% and 20.4%, respectively in terms of the number of sensor nodes in the environment with respect to the moving radius of the nodes, mobile nodes and cellular automata compared to the environment considering the constant radius of the nodes, mobile nodes and without automata.

There are many theories about the causes of hereditary diseases, but physician believe that both the genetic and environmental factors simultaneously play an important role in the development and progression of these diseases, although the extent to which this effect is not yet clear. In order to detect effective genes in the development of diseases, it is necessary to achieve the relationship between cells/tissues. The interaction between different cells/tissues can be demonstrated by expressing the gene between them. By sampling chromosomes, useful information is obtained about the type of disease and how it is transmitted. By examining this information, you can identify disorders that have led to highly altered changes. In this paper, the recognition of intercellular and inter-tissue interactions in various diseases has been done according to the characteristics of the topological structure of the graph and an improved cumulative clustering method. The proposed method has two stages; in the first step, several clustering models are combined to identify the initial relationships between cells or tissues in order to produce better results than individual algorithms. In the second stage, the similarity between cells or tissues in each cluster is calculated using a similarity criterion based on the topological structure of the graph. Eventually, the maximum similarity between cells or tissues in each cluster is used to discover the relationship between diseases. To evaluate the performance of the proposed method, several UCI datasets and the Phantom 5 dataset have been used. The results of the proposed method on the phantom data set 5 report a silhouette of 0.901 in 18 clusters for cells and 0.762 in 13 clusters for tissues.

Non-distortion-specific no-reference image quality assessment is one of the challenges in the field of digital image processing. This is because there are no reference images, the type of failure, scores, and scoring of a human observer while this field is used in various applications. The purpose of this paper is to use the properties and characteristics of the images and model them with the q-Gaussian distribution for evaluation of image quality. The q-Gaussian distribution is one of the options that creates the boundaries of flexibility decision making with different Gaussian forms that have more generalizability in abnormalities than other distributions, and to better model the statistical properties of the image. Our experimental results show that the performance of the proposed technique is more than other distribution.

Today, applications of the virtualization technology are rapidly growing in which setting up and running multiple operating systems on a single physical system. Computational clouds are the most hallmark of this technology. Intrusion detection systems play a key role in protecting cloud resources on virtual machines. Regarding the increasing speed and complexity of these machines, it is necessary to increase the ability and accuracy of intrusion detection systems to identify different types of attacks at the right time. In this regard, the use of behavior-based approaches has attracted more attention due to their high scalability in large networks. The methods for intrusion detection that utilizes network traffic graph clustering do not have the accuracy and appropriateness with the speed of data transfer in the current computer networks. Thus, the solutions can be improved by choosing an appropriate strategy for clustering. In this paper, a new behavior-based method for detecting intrusion in computer networks is presented. To this end, the network data was modeled through the flow of data as a traffic distribution graph and then clustered using an improved Markov-based algorithm. Then, the produced clusters are used to construct an intrusion detection model by analyzing a set of modified statistical criteria. The proposed model was examined and evaluated on the DARPA 99 dataset and compared with seven other robust methods. The results show that the proposed method detects attacks with high accuracy and works better than the methods which do not use the graph clustering.

Image mosaicing refers to stitching two or more images which have overlapping regions to a larger and more comprehensive image. Scale Invariant Feature transform (SIFT) is one of the most commonly used detectors previously used in image mosaicing. The defects of SIFT algorithm are the large number of redundant keypoints and high execution time due to the high dimensions of classical SIFT descriptor, that reduces the efficiency of this algorithm. In this paper, to solve these problems a new four-step approach for image mosaicing is proposed.  At first, the keypoints of both reference and sensed images are extracted based on Redundant Keypoint Elimination-SIFT (RKEM-SIFT) algorithm to improve the mosaicing process. Then, to increase the speed of the algorithm, the 64-D SIFT descriptor for keypoints description is used. Afterwards, the proposed RANdom SAmple Consensus (RANSAC) algorithm is used for removing mismatches. Finally, a new method for image blending is proposed. The details of the proposed steps are as follows. RKEM-SIFT algorithm has been proposed in [1] to eliminate redundant points based on redundancy index. In this paper, RKEM algorithm is used to extract keypoints to improve the accuracy of image mosaicing. In the second stage, for each keypoint of the image, 64-D SIFT descriptor is computed. In this descriptor, unlike the 128-D SIFT descriptor, a smaller window is used which improves the accuracy of matching and reduces the running time. In the third stage, the proposed adaptive RANSAC algorithm is suggested to determine the adaptive threshold in the RANSAC algorithm to remove the mismatches and to improve the image mosaicing. Determining the appropriate threshold value in RANSAC is so important, because if an appropriate value is not chosen for this algorithm, the mismatches are not removed, and eventually there will be a serious impact on the outcome of the image mosaicing process. In this method, the threshold value is based on the median value of distances between matching points and their transformed model. Image blending in the mosaicing process is the final step which blends the pixels intensity in the overlapped region to avoid seams. The proposed method of blending is to combine the images based on the Gaussian weighting function, which the mean of this function is considered as the average of the data in the overlapped region of two images. The proposed blending method reduces artifacts in the image for better performance of the mosaicing process. Another advantage of this proposed method is the possibility to combine more than two images that are suitable for creating panoramic images. The simulation results of the proposed image mosaicing technique, which includes the RKEM-SIFT algorithm as feature detector, 64-D SIFT descriptor, proposed adaptive RANSAC algorithm and proposed image blending algorithm on different image databases show the superiority of the proposed method according to RMSE criteria, precision and running time.

Any heart activity disorder may lead an irregularity in is rhythm, or cardiac arrhythmia. An ECG signal is one of the major tools for classifying different types of cardiac arrhythmias. ECG signals usually contain various noises. To have a better signal processing, it is essential to remove noises in a way that a signal structure never become subject to distortion. After the step of noise removal, selection of an appropriate method is of paramount importance for feature extraction. Optimal features can be selected to improve efficiency and reduce calculations.

This article used the ensemble empirical mode decomposition (EEMD) in which any intrinsic mode function (IMF) contains only a single frequency component for noise removal. The noise removal operation with the least distortion is possible using an appropriate windowing on a QRS complex containing sum of the first three IMFs. Later, the remaining noises can be removed using discrete wavelet transform (DWT). The results of using the EEMD-DWT combined method were compared with EMD and DWT combination. After the noise removal step, feature extraction was performed through a wavelet packet decomposition. It is capable of signal decomposition at all frequencies. Multiple objective binary particle swarm optimization (MOBPSO) method was used to select optimal features and the effect of this method on the results was examined. Finally, the back propagation neural network (BPNN) and a support vector machine based on particle swarm optimization were used for classification.

This article used 17 signals received from the MIT-BIH database. The acquired data belong to 6 different types of classes. After pre-processing, feature extraction, feature selection, and classification on the input data, it is observed that the proposed technique of EEMD-DWT is an appropriate method for noise removal and MOBPSO is a suitable method for the selection of best features. The BPNN classifier managed to classify cardiac arrhythmias with a higher accuracy and the values for accuracy, sensitivity, specificity, and positive predictive value were 99.12%, 97.08%, 99.38%, and 97.12%, respectively.