Vol. 41 · No. 4

This paper proposes a novel architecture that utilises an attention mechanism in conjunction with multi-stream convolutional neural networks (CNN) to obtain high accuracy in human re-identification (Reid). The proposed architecture consists of four blocks. First, the pre-processing block prepares the input data and feeds it into a spatial-temporal two-stream CNN (STC) with two fusion points that extract the spatial-temporal features. Next, the spatial-temporal attentional LSTM block (STA) automatically fine-tunes the extracted features and assigns weight to the more critical frames in the video sequence by using an attention mechanism. Extensive experiments on four of the most popular datasets support our architecture. Finally, the results are compared with the state of the art, which shows the superiority of this approach.

The paper presents a GPU implementation of the thermal discrete element method (TDEM) and the comparative analysis of its performance. Several discrete element models for granular flows, the bonded particle model and the TDEM are considered for quantitative comparison of computational performance. The performance measured on NVIDIA(R) Tesla™ P100 GPU is compared with that attained by running the same OpenCL code on Intel(R) Xeon™ E5-2630 CPU with 20 cores. The presented GPU implementation of the TDEM increases the computing time of the bonded particle model only up to 30.6

Edge detection is especially important for computer vision and generally for image processing and visual recognition. On the other hand, digital image processing is widely used in multiple science fields such as medicine, X-ray analysis, magnetic resonance tomography, computed tomography, and cosmology, i.e. information collection from satellites, its transferring, and analysis. Any step of image processing, from obtaining the image to its segmentation and object recognition is followed by image noise. The processing speed is vital in popular fields that demand image analysis in real time. In this work, we have proposed an approach of parallel computing of the Canny algorithm using CUDA technology, the complexity of object recognition is analyzed according to the type of the image noise and the level of its density. The sequenced implementation on GPU and the parallel implementation on GPU was considered. The results were analyzed for efficiency and reliability. Also, parallel acceleration is calculated according to the size of the incoming image. The manipulations with the image showed the growth of processing speed of 68 times, whereas the manipulations with the size of the kernel showed the growth of processing speed of 26 times. Another contribution of this work is the analysis of the complexity of object recognition depending on the type of image noise and the level of its density. Furthermore, the increase of Gaussian noise density linearly increases the complexity of object recognition.

One of the most challenging problems in the real-world dataset is the rising numbers of imbalanced data. The fact that the ratio of the majorities is higher than the minorities will lead to misleading results as conventional machine learning algorithms were designed on the assumption of equal class distribution. The purpose of this study is to build a hybrid data preprocessing approach to deal with the class imbalance issue by applying resampling approaches and CSL for fraud detection using a real-world dataset. The proposed hybrid approach consists of two steps in which the first step is to compare several resampling approaches to find the optimum technique with the highest performance in the validation set. While the second method used CSL with optimal weight ratio on the resampled data from the first step. The hybrid technique was found to have a positive impact of 0.987, 0.974, 0.847, 0.853 F2-measure for RF, DT, XGBOOST and LGBM, respectively. Additionally, relative to the conventional methods, it obtained the highest performance for prediction.

Graphs are widely used to model complicated data semantics in many application domains. Subgraph isomorphism checking (an NP-complete problem) is a regular operation with this kind of data. In this paper, we propose an improvement of Ullmann algorithm, a well-known subgraph isomorphism checker. Our new algorithm is called Ullmann-ONL. It utilizes a new search ordering and L-levels of vertex neighborhoods (NL) to confine the search space of Ullmann algorithm. Our performance study shows that Ullmann-ONL outperforms previously proposed algorithms with a wide margin.

In order to model complex workflow systems with databases, and detect their data-flow errors such as data inconsistency, we defined Workflow Net with Table model (WFT-net) in our previous work. We used a Petri net to describe control flows and data flows of a workflow system, and labeled some abstract table operation statements on transitions so as to simulate database operations. Meanwhile, we proposed a data refinement method to construct the state reachability graph of WFT-nets, and used it to verify some properties. However, this data refinement method has a defect, i.e., it does not consider the constraint relation between guard functions, and its state reachability graph possibly has some pseudo states. In order to overcome these problems, we propose a new data refinement method that considers some constraint relations, which can guarantee the correctness of our state reachability graph. What is more, we develop the related algorithms and tool. We also illustrate the usefulness and effectiveness of our method through some examples.

A method of repairing process models with non-free-choice constructs is proposed based on logical Petri nets, aiming at the problem of low precision in the existing repair methods. An extended successor matrix of transitions is determined according to the distance between any two transitions. There are two types of choice-construct transitions. One is a non-free-choice construct transition, and the other is a general choice construct transition. The type of choice-construct transitions can be determined based on the extended successor matrix and the relationship between the front and back sets of transitions. The location of the deviations is calculated by an improved replaying method. Finally, a model can be repaired according to remaining-token places and missing-token places. Based on the experiments on real event logs, the method proposed in this paper has a better performance in fitness, precision, and simplicity compared with its peers.

Aiming at the problem of ``dynamic'' accurate determination of rates in UBI auto insurance pricing, this paper proposes a UBI auto insurance pricing model based on fuzzy controller and optimizes it with a parameter adaptive SASPO. On the basis of the SASPO algorithm, the movement direction of the particles can be mutated and the direction can be dynamically controlled, the inertia weight value is given by the distance between the particle and the global optimal particle, and the learning factor is calculated according to the change of the fitness value, which realizes the parameter in the running process. Effective self-adjustment. A five-dimensional fuzzy controller is constructed by selecting the monthly driving mileage, the number of violations, and the driving time at night in the UBI auto insurance data. The weights are used to form fuzzy rules, and a variety of algorithms are used to optimize the membership function and fuzzy rules and compare them. The research results show that, compared with other algorithms, the parameter adaptive SAPAO algorithm can calculate more reasonable, accurate and high-quality fuzzy rules and membership functions when processing UBI auto insurance data. The accuracy and robustness of UBI auto insurance rate determination can realize dynamic and accurate determination of UBI auto insurance rates.

The COVID-19 influenza became a curse on the world. It has been around for two years, so no one needs to make a big introduction of it. It has became a significant challenge around the world. Owing to this, we made dynamic networks using an amalgamating of fuzzy logic and neural networks for the prediction of sufferers of COVID-19. These hybrid networks serve for the assessment of the COVID-19 victims and usefully serve for the assessment of the medical resources needed for future victims. This manuscript proposed Sugeno Adaptive Neuro-Fuzzy Inference System (SANFIS) prediction model for COVID-19 prediction in Andhra Pradesh, India. We gathered data on positive COVID-19 sufferers in Andhra Pradesh for this purpose. The data can be separated into three categories: training set, testing set and checking set. We have utilized Root Mean Square Deviation (RMSD) for prediction precision. If the prediction model has a lower RMSD value, it is regarded as the best forecast. In this study, we concluded that the 3 Triangular MFns for each input were excellent with the extreme precision for all of the districts based on our expertise. In the end, we deployed seven SANFIS replicas in Andhra Pradesh, but we discovered that SANFIS6 and SANFIS7 provided excellent COVID-19 prediction results. These findings will assist the government, healthcare agencies, and medical organizations in planning for future COVID-19 victims' medical requirements. These sorts of Sugeno Adaptive Neuro-Fuzzy Inference System (SANFIS) prediction models based on Artificial Intelligence (AI) will be beneficial in overcoming the COVID-19.

3D recovery approaches require a variety of clues to obtain shape information. The shape from shading (SFS) method uses shading variations in images to estimate depth maps. Although shading contains detailed information, it causes some well-known ambiguities such as convex-concave ambiguity. In this study, a system installation using red, green, and blue illumination and an algorithm processing reflections on the surface were proposed to accurately analyze surface orientations and solve ambiguity problems. The algorithm evaluated combinations of light hitting the surface from different directions and detailed surface orientations to avoid erroneous predictions. The proposed system was tested with eight different methods in the literature developed from the earliest times to the present, and the initially erroneously predicted surface orientations were improved. Consequently, the correct orientation of the surface points was determined by removing the ambiguities in images taken without considering the location of illumination, and all the tested methods provided successful results using the proposed system.