JOURNAL OF ARTIFICIAL INTELLIGENCE AND DATA MINING
Year:2021 | Volume:9 | Issue:1|Papers Count:11

In recent years, vehicle classification has been one of the most important research topics. However, due to the lack of a proper dataset, this field has not been well-developed as other fields of intelligent traffic management. Therefore, the preparation of large-scale datasets of vehicles for each country is of great interest. In this paper, we introduce a new standard dataset of popular Iranian vehicles. This dataset, which consists of the images of the vehicles moving in urban streets and highways, can be used for vehicle classification and license plate recognition. It contains a large collection of vehicle images in different weather and lighting conditions with different viewing angles. It took more than a year to construct this dataset. The images were taken from various types of mounted cameras with different resolutions and at different altitudes. In order to estimate the complexity of the dataset, some classical methods alongside the popular deep neural networks were trained and evaluated on the dataset. Furthermore, two light-weight CNN structures are also proposed, one with three Conv layers and the other with five Conv layers. The 5-Conv model with 152K parameters reached the recognition rate of 99. 09% and could process 48 frames per second on CPU, which is suitable for real-time applications.

Estimation of the blurriness value in an image is an important issue in the image processing applications such as image deblurring. In this paper, a no-reference blur metric with a low computational cost is proposed, which is based on the difference between the second-order gradients of a sharp image and the one associated with its blurred version. The experiments, in this work, are performed on four databases including CSIQ, TID2008, IVC, and LIVE. The experimental results obtained indicate the capability of the proposed blur metric in measuring image blurriness and also the low computational cost compared with the other existing approaches.

Periodic noise reduction is a fundamental problem in image processing, which severely affects the visual quality and the subsequent application of the data. Most of the conventional approaches are only dedicated to either the frequency or the spatial domain. In this research work, we propose a dual-domain approach by converting the periodic noise reduction task into an image decomposition problem. We introduce a bio-inspired computational model to separate the original image from the noise pattern without having any a priori knowledge about its structure or statistics. Experiments on both the synthetic and non-synthetic noisy images are carried out in order to validate the effectiveness and efficiency of the proposed algorithm. The obtained results demonstrate the effectiveness of the proposed method both qualitatively and quantitatively.

Face recognition is a challenging problem due to different illuminations, poses, facial expressions, and occlusions. In this paper, a new robust face recognition method is proposed based on the color and edge orientation difference histogram. Firstly, the color and edge orientation difference histogram is extracted using color, color difference, edge orientation, and edge orientation difference of the face image. Then the backward feature selection is employed in order to reduce the number of features. Finally, the Canberra measure is used to assess the similarity between the images. The color and edge orientation difference histogram shows the color and edge orientation difference between two neighboring pixels. This histogram is effective for face recognition due to the different skin colors and different edge orientations of the face image, which leads to a different light reflection. The proposed method is evaluated on the Yale and ORL face datasets. These datasets consist of gray-scale face images under different illuminations, poses, facial expressions, and occlusions. The recognition rate over the Yale and ORL datasets is achieved to be 100% and 98. 75%, respectively. The experimental results demonstrate that the proposed method outperforms the existing methods in face recognition.

Today, feature selection, as a technique to improve the performance of the classification methods, has been widely considered by the computer scientists. As the dimensions of a matrix has a huge impact on the performance of processing on it, reducing the number of features by choosing the best subset of all the features. It will affect the performance of the algorithms. Finding the best subset by comparing all the possible subsets, even when n is small, is an intractable process, and hence, many research works have approached to the heuristic methods to find a near-optimal solutions. In this paper, we introduce a novel feature selection technique that selects the most informative features and omits the redundant or irrelevant ones. Our method is embedded in PSO (Particle Swarm Optimization). In order to omit the redundant or irrelevant features, it is necessary to figure out the relationship between different features. There are many correlation functions that can reveal this relationship. In our proposed method, to find this relationship, we use the mutual information technique. We evaluate the performance of our method on three classification benchmarks: Glass, Vowel, and Wine. Comparing the results obtained with four state-of-the-art methods demonstrates its superiority over them.

Social networks are valuable sources for the marketers, who can publish campaigns to reach target audiences according to their interest. Although Telegram was primarily designed as an instant messenger, it is now used as a social network in Iran due to the censorship of Facebook, Twitter, etc. Telegram neither provides a marketing platform nor the possibility to search among groups. It is difficult for the marketers to find target audience groups in Telegram, and hence, we have developed a system to fill the gap. The marketers use our system to find target audience groups by keyword search. Our system has to search and rank groups as relevant as possible to the search query. This paper proposes a method called GroupRank to improve the ranking of group searching. GroupRank elicits associative connections among groups based on membership records they have in common. After a detailed analysis, five group quality factors are introduced and used in the ranking. Our proposed method combines the TF-IDF scoring with group quality scores and associative connections among groups. The experimental results show improvement in many different queries.

The SailFish Optimizer (SFO) is a metaheuristic algorithm inspired by a group of hunting sailfish that alternate their attacks on a group of prey. The SFO algorithm takes advantage of using a simple method for providing a dynamic balance between the exploration and exploitation phases, creating the swarm diversity, avoiding local optima, and guaranteeing a high convergence speed. Nowadays, multi-agent systems and metaheuristic algorithms can provide high performance solutions for solving combinatorial optimization problems. These methods provide a prominent approach to reduce the execution time and improve the solution quality. In this paper, we elaborate a multi-agent based and distributed method for sailfish optimizer (DSFO), which improves the execution time and speeds up the algorithm, while maintaining the optimization results in a high quality. The Graphics Processing Units (GPUs) using Compute Unified Device Architecture (CUDA) are used for the massive computation requirements in this approach. In depth of the study, we present the implementation details and performance observations of the DSFO algorithm. Also a comparative study of the distributed and sequential SFO is performed on a set of standard benchmark optimization functions. Moreover, the execution time of the distributed SFO is compared with other parallel algorithms to show the speed of the proposed algorithm to solve the unconstrained optimization problems. The final results indicate that the proposed method is executed about maximum 14 times faster than the other parallel algorithms and shows the ability of DSFO for solving the non-separable, non-convex, and scalable optimization problems.

The Internet of Things (IoT) is a novel paradigm in computer networks is capable of connecting things to the internet via a wide range of technologies. Due to the features of the sensors used in the IoT networks and the unsecured nature of the internet, IoT is vulnerable to many internal routing attacks. Using the traditional IDS in these networks has its own challenges due to the resource constraint of the nodes and the characteristics of the IoT network. A sinkhole attacker node in this network attempts to attract traffic through an incorrect information advertisement. In this research work, a distributed IDS architecture is proposed in order to detect the sinkhole routing attack in the RPL-based IoT networks, and this is aimed to improve a true detection rate and reduce the false alarms. For the latter, we used one type of post-processing mechanism in which a threshold is defined for separating suspicious alarms for further verifications. Also the implemented IDS modules are distributed via the client and router border nodes that make it energy efficient. The required data for interpretation of the network’, s behavior is gathered from the scenarios implemented in the Cooja environment with the aim of Rapidminer for mining the produced patterns. The produced dataset is optimized using the genetic algorithm by selecting appropriate features. We investigate three different classification algorithms, and in its best case, Decision Tree could reach a 99. 35 rate of accuracy.

Aspect-level sentiment classification is an essential issue in the sentiment analysis that intends to resolve the sentiment polarity of a specific aspect mentioned in the input text. The recent methods have discovered the roles of some aspects in sentiment polarity classification and have developed various techniques to assess the sentiment polarity of each aspect in the text. However, these studies do not pay enough attention to the need for vectors to be optimal for the aspects. In order to address this issue, in the present work, we suggest a Hierarchical Attention-based Method (HAM) for the aspect-based polarity classification of the text. HAM works in a hierarchically manner. Firstly, it extracts an embedding vector for the aspects. Next, it employs these aspect vectors with information content to determine the sentiment of the text. The experimental findings on the SemEval2014 dataset show that HAM can improve the accuracy by up to 6. 74% compared to the state-of-the-art methods in the aspect-based sentiment classification task.

the developers of complex business information systems to expose the configuration parameters to the system administrators. This allows them to intervene by tuning the bottleneck configuration parameters in response to the current changes or in anticipation of the future changes in order to maintain the system performance at an optimum level. However, these manual performance tuning interventions are prone to error and lack of standards due to fatigue, varying levels of expertise, and over-reliance on inaccurate predictions of future states of a business information system. The purpose of this research work is to investigate that how the capacity of probabilistic reasoning to handle uncertainty can be combined with the capacity of Markov chains to map the stochastic environmental phenomena to ideal self-optimization actions. This is done using a comparative experimental research design that involves quantitative data collection through simulations of different algorithm variants. This provided compelling results, which indicate that applying the algorithm to a distributed database system improves the performance of tuning decisions under uncertainty. The improvement is measured quantitatively by a response-time latency 27% lower than the average and a transaction throughput 17% higher than the average.

Scouring, occurring when the water flow erodes the bed materials around the bridge pier structure, is a serious safety assessment problem for which there are many equations and models available in the literature in order to estimate the approximate scour depth. This research work is aimed to study how the surrogate models estimate the scour depth around circular piers, and compare the results with those of the empirical formulations. To this end, the pier scour depth is estimated in non-cohesive soils based on a sub-critical flow and live bed conditions using the artificial neural networks (ANNs), group method of data handling (GMDH), multivariate adaptive regression splines (MARS), and Gaussian process models (Kriging). A database containing 246 lab data gathered from various studies is formed, and the data is divided into three random parts: 1) training, 2) validation, and 3) testing in order to build the surrogate models. The statistical error criteria such as the coefficient of determination (R2), root mean squared error (RMSE), mean absolute percentage error (MAPE), and absolute maximum percentage error (MPE) of the surrogate models are then found and compared with those of the popular empirical formulations. The results obtained reveal that the surrogate models‘,test data estimations are more accurate than those of the empirical equations,Kriging has better estimations than the other models. In addition, the sensitivity analyses of all the surrogate models show that the pier width‘, s dimensionless expression (b/y) has a greater effect on estimating the normalized scour depth (Ds/y).