فصلنامه پردازش علائم و داده ها
سال نوزدهم شماره 4 (پیاپی 54، زمستان 1401)

In this manuscript, we proposed a new scheme in communication signal detection which is respect to the curve shape of received signal and based on the extraction of curve fitting (CF) features. This feature extraction technique is proposed for signal data classification in receiver. The proposed scheme is based on curve fitting and approximation of rational fraction coefficients. For each symbol of received signal, a specific rational function approximation is developed to fit with received signal curve and the coefficients of the numerator and denominator polynomials of this function are considered as new extracted features. Then  it will be shown that the coefficients of this polynomials have the potential for using as new features in a statistical classifier and have better performance in competition with other solutions such as linear and even nonlinear feature extraction methods in  classification. The criteria used in performance evaluation are  probability of error and signal to noise ratio in FSK and ASK modulations. For each symbol of received signal, a specific rational function approximation is developed to fit with received signal curve and the coefficients of the numerator and denominator polynomials of this function are considered as new extracted features. In the proposed method, there are two phases train and test, which are described in the following two steps. First, in the train phase, the algorithm starts by using binary FSK and ASK modulations, so first, a system generate a number of random symbols then signal is modulated by binary ASK and FSK. The Modulated FSK and ASK signals are corrupted in the channel with noise. The noise-corrupted signal enters the receiver at the corresponding transmitted interval. Then, the samples are extracted from the modulated signals based on predetermined sample rates. Then, we fit a rational fraction curve with degrees of L and M to each set of N samples. Afterward, we apply all the numerator (L+1) and denominator (M) coefficients to 0 and 1 classes  in the binary FSK and ASK modulations. We store all the specific coefficients of the deterministic symbols at different M and L values to create the corresponding histogram in each class. In each histogram (i.e., the coefficients of a class), we extract and store specific coefficients that completely discriminate between the two classes. Therefore, we determine all the histograms where there is a good approximation of discrimination and create the related table. Note that the data used in histograms are the training data. Then, in order to analyze and evaluate the performance of the proposed curve fitting method, we send the testing data through the channel corresponding to the related modulator. The results of the proposed classification method show that it provides smaller error rate regarding to the theoretical error rate probability in AWGN channel.

Feature selection is considered as an important issue in classification domain. Selecting a good feature through maximum relevance criterion to class label and minimum redundancy among features affect improving the classification accuracy. However, most current feature selection algorithms just work with the centralized methods. In this paper, we suggest a distributed version of the mRMR feature selection approach. In mRMR, feature selection is performed based on maximum relevance to class and minimum redundancy among the features. The suggested method include six stages: in the first stage, after determining training and test data, training data are distributed horizontally. All subsets have same number of features. In the second stage, each subset of features is scored using mRMR feature selection. Features with higher ranks are selected and others are eliminated. In the fourth stage, features which were omitted are voted. In the fifth stage, the selected features are merged to determine the final set. In the final stage, classification accuracy is evaluated using final training data and test data. Our method quality has been evaluated by six datasets. The results prove that the suggested method can improve classification accuracy compared to methods just based on maximum relevance to class label in addition to runtime reduction.

This research is related to the use of deep learning tools and image processing technology in the automatic generation of images from text. Previous researches have used one sentence to produce images. In this research, a memory-based hierarchical model is presented that uses three different descriptions that are presented in the form of sentences to produce and improve the image. The proposed scheme focuses on using more information to produce high-resolution images, using competitive productive networks. Implementing programs related to this field require massive processing resources. Therefore, the proposed method was implemented and tested on a cluster with 25 GPUs using the hardware platform of the University of Copenhagen. The experiments were performed on CUB-200 and ids-ade datasets. The experimental results show that the proposed model can produce higher quality images than the two basic models StackGAN and AttGAN.

Unemployment insurance is one of the most popular insurance types in the modern world. The Social Security Organization is responsible for checking the unemployment benefits of individuals supported by unemployment insurance. Hand-crafted evaluation of unemployment claims requires a big deal of time and money. Data mining and machine learning as two efficient tools for data analysis can assist Social Security Organization in automating this process. In this research work, a hybrid supervised learning method is proposed to verify the eligibility of applicants for unemployment. The proposed method takes as input the information of insured individuals, and assigns a numeric score to each applicant through analyzing the input data. Then, claimants are classified into two groups according to those scores: "Qualified” and "Unqualified". The proposed method includes two hybrid strategies: BSA-SVM and combination of confidence values. In BSA-SVM method, backtracking search algorithm (BSA) is used to estimate the prameters of support vector machines (SVM) and improves the classification performance. In the second approach, confidence values extracted from individual classofiers are combined to better classify the input data. Empirical evaluation shows an accuracy of 87% for BSA-SVM and 86% for the second approach.

Today, data plays an essential role in all levels of human life, from personal cell phones to medical, educational, military and government agencies. In such circumstances, the rate of cyber-attacks is also increasing. According to official reports, data breaches exposed 4.1 billion records in the first half of 2019. An information system consists of several components, which one of the most important them is the database. A database in addition to being a repository of data, acts as a common information bus between system components. For this reason, any attack on the database may disrupt the operation of other components of the system. In fact, database security is shared throughout the whole information system. The attack may carried out in various ways, such as data theft, damaging data, and privacy breach. According to the sensitivity of the stored data, database attack could lead to significant human and financial losses even at the national level. Among the different types of threats, since legitimate operator plays a key role in an information system, his/her threat is one of the most dangerous threats to the security and integrity of a database system. This type of cyber-attack occurs when an insider operator abuses his/her legal permissions in order to access unauthorized data. In this paper, a new performance-based authorization framework has been presented which is able to reduce the potential of insider threat in the database system. The proposed method insure that only authenticated operator performs authorized activities on the database objects. In the proposed framework, the access permission of the operator to a database table is determined using his/her performance and the level of sensitivity of the table. The value of the operator performance is updated periodically or when an abuse is detected, in order to protect access to the contents of a database as well as preserve the consistency, integrity, and overall quality of the data. Simulation results, using real dataset from a hospital information system, indicate that the proposed framework has effective performance for mitigating insider threats.

Nowadays, achieving desirable and stable security in networks with national and organizational scope and even in sensitive information systems, should be based on a systematic and comprehensive method and should be done step by step. Cryptography is the most important mechanism for securing information. a cryptographic system consists of three main components: cryptographic algorithms, cryptographic keys, and security protocols, which are mainly based on cryptographic algorithms. In designing a cryptographic algorithm, all the necessary components of information security must be considered in a model of excellence in technical, organizational, procedural and human aspects. To meet these needs, we must first extract the effective components in the design and implementation of cryptographic algorithms based on a model and then determine the impact of the components. In this paper, we use cybernetic methodology to prepare a   metamodel.The cryptographic cybernetics metamodel has four components: " strategy / policy ", "main process", "support process" and "control process". The "main process" has four stages and also, the "suport process" includes 13 components of hardware and software. The interactions of these two processes shape its structure, leading to a complex graph. To prioritize suport components for resource allocation and cryptography strategy, it is necessary to rank these components in the designed metamodel. To overcome this complexity in order to rank the support components, we use the ELECTRE III method, which is a multi-criteria decision-making method. The results show that the components with high priority for the development of the cryptographic system are: Research and Development, Human Resources, Management, Organizational, Information and Communication Technology, Rrules and Regulations and standards. These results are consistent with reports published by the ITU in 2015, 2017 and 2018.

Over the last years‎, ‎the concept of Internet of Things (IoT) leads to a revolution in the communications of humans and things. ‎Security and efficiency could be the main challenges of that communication‎‎. ‎‎On the other hand, ‎ authenticity and confidentiality are two important goals to provide desired security in an information system‎, including IoT-based applications. An Authentication and Key Agreement (AKA) protocol is a tool to ‎achieve authenticity and agree on a secret key to reach confidentiality. Therefor using a secure AKA protocol, one can establish the mentioned security. ‎In the last years‎, ‎several articles have discussed AKA protocols in the WSN‎. ‎For example‎, ‎in 2014‎, ‎Turkanovic et al‎. proposed a new AKA scheme for the heterogeneous ad-hoc WSN. ‎In 2016‎, ‎Sabzinejad et al‎. presented an improved one. ‎In 2017‎, ‎Jiang et al‎. introduced a secure AKA protocol‎. ‎Some other AKA protocols have presented in the last three years. ‎All the mentioned protocols are lightweight ones and need minimum resources and try to decrease the computation and communication costs in the WSN context‎. ‎In 2019‎, ‎Janababaei et al. proposed an AKA scheme in the WSN for the IoT applications, in the journal of Signal and Data Processing (JSDP)‎. ‎In the context of efficiency‎, ‎the protocol only uses a hash function‎, ‎bitwise XOR‎, ‎and concatenation operation‎. ‎Hence‎, ‎it can be  considered as a lightweight protocol‎. ‎The authors also discussed the security of their scheme and claimed that the proposed protocol has the capability  to offer anonymity and trust and is secure against traceability‎, ‎impersonation‎, ‎reply and man in the middle attacks‎. ‎However, despite their claims‎, ‎this research highlights some vulnerabilities in that protocol, for the first time to the best of our knowledge‎. More precisely, we showe that a malicious sensor node can find the secret parameters of another sensor node when it establishes a session with the victimized sensor. Besides, an adversary can determine any session key of two sensor nodes, given only a known session key of them. We also show that the protocol could not satisfy the anonymity of the sensor nodes. Other attacks which influence the Janababaei et al.’s scheme, are impersonation attack on the sensor nodes and cluster heads and also the man in the middle attack. In this paper we find that the main weaknesses of the Janababaei et al.’s protocol are related to computation of the session key, . We also propose a simple remedy to enhance the security of the Janababaei et al.’s protocol. ‎An initial attempt to improve the protocol is using a hash function on the calculated key, . This suggestion is presented to enhance the security of the protocol against the observed weaknesses in this paper; but it does not mean that there are no other security issues in the protocol. Therefore, modification and improvement of the Janababaei et al.’s protocol such that it provides other security features can be considered in the future research of this paper. Besides, since in this paper we focus on the security of the protocol, then the efficiency of it was not discussed. Therefore one can consider the modification of the message structure of the protocol to reduce the computational and telecommunication costs of it as another future work in the context of this paper.

Data clustering is one of the main steps in data mining, which is responsible for exploring hidden patterns in non-tagged data. Due to the complexity of the problem and the weakness of the basic clustering methods, most studies today are guided by clustering ensemble methods. Diversity in primary results is one of the most important factors that can affect the quality of the final results. Also, the quality of the initial results is another factor that affects the quality of the results of the ensemble. Both factors have been considered in recent research on ensemble clustering. Here, a new framework for improving the efficiency of clustering has been proposed, which is based on the use of a subset of primary clusters, and the proposed method answers the above questions and ambiguities. The selection of this subset plays a vital role in the efficiency of the assembly. Since evolutionary intelligent algorithms have been able to solve the majority of complex engineering problems, this paper also uses these intelligent methods to select subsets of primary clusters. This selection is done using three intelligent methods (genetic algorithm, simulation annealing and particle swarm optimization). In this paper a clustering ensemble method is proposed which is based on a subset of primary clusters. The main idea behind this method is using more stable clusters in the ensemble. The stability is applied as a goodness measure of the clusters. The clusters which satisfy a threshold of this measure are selected to participate in the ensemble. For combining the chosen clusters, a co-association based consensus function is applied. A new EAC based method which is called Extended Evidence Accumulation Clustering, EEAC, is proposed for constructing the Co-association Matrix from the subset of clusters. Experimental results on several standard datasets with normalized mutual information evaluation, Fisher and accuracy criteria compared to Alizadeh, Azimi, Berikov, CLWGC, RCESCC, KME, CFSFDP, DBSCAB, NSC and Chen methods show the significant improvement of the proposed method in comparison with other ones.

As we know, credit cards speed up and make life easier for all citizens and bank customers. They can use it anytime and anyplace according to their personal needs, instantly and quickly and without hassle, without worrying about carrying a lot of cash and more security than having liquidity. Together, these factors make credit cards one of the most popular forms of online banking. This has led to widespread and increasing use for easy payment for purchases made through mobile phones, the Internet, ATMs, and so on. Despite the popularity and ease of payment with credit cards, there are various security problems, increasing day by day. One of the most important and constant challenges in this field is credit card fraud all around the world. Due to the increasing security issues in credit cards, fraudsters are also updating themselves. In general, as a field grows in popularity, more fraudsters are attracted to it, and this is where credit card security comes into play. So naturally, this worries banks and their customers around the world. Meanwhile, financial information acts as the main factor in market financial transactions. For this reason, many researchers have tried to prioritize various solutions for detecting, predicting, and preventing credit card fraud in their research work and provide essential suggestions that have been associated with significant success. One of the practical and successful methods is data mining and machine learning. In these methods, one of the most critical parameters in fraud prediction and detection is the accuracy of fraud transaction detection. This research intends to examine the Gradient Boosting methods, which are a subset of Ensemble Learning and machine learning methods. By combining these methods, we can identify credit card fraud, reduce error rates, and improve the detection process, which in turn increases efficiency and accuracy. This study compared the two algorithms LightGBM and XGBoost, merged them using simple and weighted averaging techniques, and then evaluate the models using AUC, Recall, F1-score, Precision, and Accuracy. The proposed model provided 95.08, 90.57, 89.35, 88.28, and 99.27, respectively, after applying feature engineering and using the weighted average approach for the mentioned validation parameters. As a result, function engineering and weighted averaging significantly improved prediction and detection accuracy.

Text classification is one of the main tasks of natural language processing (NLP). In this task, documents are classified into pre-defined categories. There is lots of news spreading on the web. A text classifier can categorize news automatically and this facilitates and accelerates access to the news. The first step in text classification is to represent documents in a suitable way that can be distinguishable by a classifier. There is an abundance of methods in the literature for document representation which can be divided into a bag of words model, graph-based methods, word embedding pooling, neural network-based, and topic modeling based methods. Most of these methods only use local word co-occurrences to generate document embeddings. Local word co-occurrences miss the overall view of a document and topical information which can be very useful for classifying news articles.  In this paper, we propose a method that utilizes term-document and document-topic matrix to generate richer representations for documents.  Term-document matrix represents a document in a specific way where each word plays a role in representing a document. The generalization power of this type of representation for text classification and information retrieval is not very well. This matrix is created based on global co-occurrences (in document-level). These types of co-occurrences are more suitable for text classification than local co-occurrences. Document-topic matrix represents a document in an abstract way and the higher level co-occurrences are used to generate this matrix. So this type of representation has a good generalization power for text classification but it is so high-level and misses the rare words as features which can be very useful for text classification. The proposed approach is an unsupervised document-embedding model that utilizes the benefit of both document-topic and term-document matrices to generate a richer representation for documents. This method constructs a tensor with the help of these two matrices and applied tensor factorization to reveal the hidden aspects of data. The proposed method is evaluated on the task of text classification on 20-Newsgroups and R8 datasets which are benchmark datasets in the news classification area. The results show the superiority of the proposed model with respect to baseline methods. The accuracy of text classification is improved by 3%.

Epilepsy is a chronic disorder of brain function caused by abnormal and excessive electrical neurons discharge in the brain. Seizures cause disturbances in consciousness that occur without prior notice, so their prediction ability, based on EEG data, can reduce stress and improve quality of life. An epileptic patient EEG data consists of five parts: Ictal, Inter-Ictal, pre-Ictal, Post-Ictal, and IT (seconds before Ictal onset). The purpose of predicting an attack is to detect the period of pre-ictal or IT to create warnings for medical procedures that are actually determined hours or minutes before ictal and do not necessarily mean the exact time of ictal [4]. The aim of many studies has been to identify the pre-ictal period based on EEG data. However, the problem of reliable prediction of epileptic seizures remains largely unsolved [5].  EEG and IEEG data types are used in detection and predicting methods. Due to the fact that artifacts and noises have a greater effect on EEG than IEEG, if there is IEEG, it has been tried to use it [6, 7]. Seizure warning methods that have a clinical application are generally based on the use on EEG [8]. Numerous studies have been performed to detect and predict seizures. The methods of signal processing and feature extraction are same in detection and prediction, but the difference is that, in detection, ictal and inter-ictal periods are compared, while in prediction, pre-ictal or IT and inter-ictal periods are being compared. Some algorithms use data modeling to extract features. References [13, 14], the coefficients AR model for the EEG data is obtained with least squares estimator, then the model coefficients are classified by SVM binary classification. In the article [15] the non-Gaussian EEG is considered using the ARIMA model (Autoregressive integrated moving average). In references [16, 17], predictions are performed based on the dynamic model with hidden variable and the sparse LVAR model, respectively. Also other features such as Mean Phase Coherency [18-20], Lag Synchronization Index to compare phase Synchronization between irregular oscillations [8,21], eigenspectra of space-delay correlation and covariance matrices [22], Largest Lyapunov Exponent [23, 25], decorrelation time, Hjorth parameters such as mobility and complexity, power spectrum in frequency bands, spectral edge frequency, the four statistical moments: mean, variance, kurtosis, skewness and there are features based on entropy and probability [6, 26-29]. Empirical mode decomposition (EMD) and wavelet transform methods have also been used to extract the feature [2, 30, 31, 37]. In articles [32, 33], the Cepstrum method has been used on short time multi channels EEG and IEEG in different patient states. Cepstrum is used to extract slow and periodic changes in speech that can be used to detect the ictal period from the inter-ictal, and has also been used to linearize the EEG [34]. In the paper [33], Cepstrum coefficients of multi-channel EEG are calculated and the 9 first coefficients are considered, then calculates the velocity and acceleration of the desired coefficients and uses a neural network to detect an epileptic seizure. The method of this paper was improved in 2014. In this way, first the signal energy and coefficients of Cepstrum are calculated and then the same process is followed. The accuracy values ​​of velocity and acceleration coefficients in this study were 89.7% - 98.7% and 98.9% - 99.9%, respectively [32].  In this study, the period of IT was detected in patients with temporal lobe epilepsy (TLE), which is the most common type of epilepsy [38]. For this purpose, two long term EEG channels LTM (long term monitoring) with a sampling rate 256, which are facing each other have been used. First, the desired signal is considered by the moving window with a length if 5 seconds and 80% overlap. The desired signal is normalized and its linear trend is removed and band-pass filtered (220 order FIR filter, cutoff at 6-20 Hz). Then the filtered date will de decomposed using discrete wavelet transform with 6-levels and Daubechies4 mother wavelet. In this step we will have 12 outputs. Next, by windowing of 500 samples and 75% overlap, the AR model with 8 order is applied to outputs. Cepstrum method can be used to detect regular and periodic changes in the ictal period of the EEG signal. According to this feature, the Cepstrum coefficients of the data window are calculated and the first coefficient of each window is considered. By applying a median filter to the 12 outputs of the previous stage, the current period of the first channel is compared to the background period of the same channel and the second channel, and the same is done for the second channel. This method reduces artifact error and inter-attack discharges. Finally, the signal is averaged by the moving window and the positive envelope of the curve is calculated. Given that we will eventually have 12 outputs, 12 threshold values are obtained for a patient’s training data, then these values are checked on the test data.  The proposed method was reviewed on a proposed model of adult epilepsy as well as 10 patients with long-term EEG data without artifact removal. Accuracy and average prediction time were 92% and 18.5 seconds, respectively. The algorithm performed better than other methods. Another advantage of the algorithm is the ability to reduce artifacts, while many studies have used short-term data without artifacts. Artifacts are located at different frequencies, which frequency analysis is performed by wavelet transform. Because two channel artifacts are unequal at the same time and in the same channel at different times, the artifacts are reduced by comparing the channels to each other. Algorithm testing on more patients is recommended to confirm the performance of the algorithm clinically.

Speech recognition is a subfield of artificial intelligence that develops technologies to convert speech utterance into transcription. So far, various methods such as hidden Markov models and artificial neural networks have been used to develop speech recognition systems. In most of these systems, the speech signal frames are processed uniformly, while the information is not evenly distributed in all of them. Auditory experiments have also shown that the human brain pays more attention to information-rich areas. By focusing on these areas instead of uniform processing, the brain can more robustly recognize speech in intrinsic and environmental speech variations such as speaker and noise. In contrast, the performance of most speech recognition systems degrades dramatically in these conditions. Therefore, to boost speech recognition systems' robustness, some researchers have focused on developing speech recognition systems by modeling these informative parts of the speech signal named landmarks. Similarly, in this article, we implemented a landmark-based system to obtain a robust Persian speech recognition system inspired by human brain perception. We also conducted neural networks-based variation compensation methods to boost its performance. In this article, acoustic landmarks are classified into two categories of events and states with the following definitions. Events are defined as areas of the speech signal in which the spectral characteristics change drastically while their length does not change a lot. The transition areas between some adjacent pairs of phones (phones' borders) are primarily selected as events. States are also defined as areas of the speech signal that spectral characteristics do not change significantly. Here the nuclei of phones are considered as the states. Previous research, linguistic sources, and implementation results have been used to determine the Persian language's appropriate landmarks. Finally, a set of 313 landmarks was selected and used in our acoustic landmarks-based phone recognition system.  The neural network structure used to recognize acoustic landmarks is a feed-forward fully connected structure with ReLU function in its hidden layers and a linear function in its final layer. The number of layers and neurons of this structure has been determined experimentally. The best structure is composed of 5 fully connected layers with 1000 neurons per layer. In this study, instead of considering 313 neurons to express each of the 313 landmarks, a heuristic labeling method is used to reduce the number of output neurons and utilize the shared information between the landmarks. The landmark recognition model slides on the speech feature sequence in the test phase to produce the output landmark sequence. Finally, to convert the obtained landmark sequence to a phone sequence, three rule-based post-processing steps are performed.  Variabilities are among the essential quality degradation sources in speech recognition; therefore, we proposed two approaches to reduce them and boost phone recognition quality in our landmark-based system. To this aim, we have utilized the nonlinear filtering characteristic of neural networks by implementing four neural network schemes. In scheme 1, a feed-forward neural network is first trained to map training landmarks to their corresponding well-recognized samples. Then this structure can act as a nonlinear filter before the landmark recognition block. In scheme 2, a unified structure is simultaneously trained to learn landmark labels and the filtering part. In both of these schemes, we used a recursive loop to increase the chance of attractor manipulation in the structures. In scheme 3, a recursive loop is added to one hidden layer. This loop acts as an input variability simulator and forces the network to recognize the input data and its variations correctly. Finally, in scheme four, a deep attractor neural network-based structure is proposed to shape the structure’s hidden layer components so that it can compensate for variabilities. The experiments are implemented on a Persian database named Farsdat, and the results are reported using phone error rate (PER) criteria. From every 25-millisecond speech frame, an acoustic feature called LHCB is extracted and combined with delta and delta-delta features of that frame. Every frame's features are concatenated with fourteen adjacent frames and are finally fed to our neural network-based landmark extraction model. The best-trained model obtained the PER of 21.74% on test data. Using scheme one to four, we achieved an absolute PER decrease by 0.39, 0.58, 0.43 and 1.30 percent, respectively. Comparing our landmark-based system's performance with other Persian phone recognition systems shows that this method could perform efficiently as a Persian phone recognition system.  In our future works, we intend to compare our acoustic-based phone recognition system's performance with conventional methods such as CTC in noisy conditions. Besides, it seems that acoustic landmarks can be used to create an alignment of the input speech sequence and the output transcription. Therefore, we will present a combination of CTC-based methods and acoustic landmarks to utilize acoustic landmarks' complementary information. This information might boost the performance and speed of CTC-based speech recognition methods, particularly in low resource languages.