2021

Browse

Filters

202127

Settings

Search Results

Now showing 1 - 10 of 27
  • Thumbnail Image
    Item
    CSE Automation System
    (Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur, Bangladesh, 2021-03-30) Mounkambou, Abdel Karim; Amadou, Hayatou
    The COVID-19 pandemic has affected educational systems worldwide, leading to the near-total closures of schools, universities and colleges. Most governments decided to temporarily close educational institutions in an attempt to reduce the spread of COVID-19. As of 12 January 2021, approximately 825 million learners are currently affected due to school closures in response to the pandemic. According to UNICEF monitoring, 23 countries are currently implementing nationwide closures and 40 are implementing local closures, impacting about 47% of the world's student population. Based on these facts we intended to provide a web platform on which teachers and students can interact without a risk to attract COVID-19. We used a model view controller (MVC) architecture and various web technologies such as html, CSS and JavaScript for front-end, and php for back-end using Laravel framework. The use of this platform can make significant contribution in helping, students to engage, to plan, execute, and assess a specific learning process.
  • Thumbnail Image
    Item
    Intrusion detection in IoT based systems
    (Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur, Bangladesh, 2021-03-30) Boubakari, Abdoul Bagui; Gelany, Aly Abdelkader; Yousufzai, Abdul Aziz; Harir, Hamdan
  • Thumbnail Image
    Item
    Optimized Human-Emotion Detection in Written-Text using Hybrid Machine Learning Classification Algorithm
    (Department of Computer Science and Engineering (CSE), Islamic University of Technology (IUT), Board Bazar, Gazipur-1704, Bangladesh, 2021-03-30) Olabi, Fopa Yuffon Amadou; Moctar, Mohamadou; Namba, Mikayilou
    No part of our psychological life is more essential to the quality and significance of our reality than emotions. In psychology, Emotion is often defined as a complex state of feeling that results in physical and psychological changes that influence thought and behavior. Emotionality is associated with a range of psychological phenomena, including temperament, personality, mood, and motivation. In 1972, psychologist Paul Eckman suggested that six basic emotions are universal throughout human cultures: fear, disgust, anger, surprise, happiness, and sadness. Emotion Recognition is an important area of work to improve the interaction between humans and machines. Emotion Detection will play a promising role in the field of Artificial Intelligence, especially in the case of Human-Machine Interface Development, Human-Computer Interaction (HCI), User-Experience (UX), and Designs. In our study case, we went through the vast area of Emotion Recognition and Detection from an AI and ML perspective, in which different parameters were taken into consideration. In this work, through our research, we developed a Human-Emotion Detection methodology based on Written-Text using a preprocessing technique based on meaningless stop words removal and a Hybrid-ML Algorithm, which is made of a Naïve-Bayes Classifier (NBC) and a Convolutional Neural Network (CNN) for a better accuracy alongside with an Optimized Text- Analysis method for Preprocessing. The preprocessing is built up around many different techniques that help the data to be reliable, standardized, and clean. It all started with the stop word removal which is one of the key parts of our work, then the standardization of the data and the following part was tokenization, followed by the TF-IDF Vectorization which was applied and we finished by a vocabulary construction.
  • Thumbnail Image
    Item
    Medical Expertise Style Transfer using Denoising Autoencoder
    (Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur, Bangladesh, 2021-03-15) Irbaz, Mohammad Sabik; Azad, Abir; Preoty, Anika Tasnim; Shalanyuy, Tani Barkat
    Due to the huge cognitive bias and the curse of knowledge, there is a notable communication gap between experts and laymen. This communication gap creates a huge problem in the medical domain. The patients do not understand what the doctors (domain expert) are saying and the doctors also face some ambiguity issues since they are not used to the laymen style. Bridging the gap between laymen and experts is a challenging task as it requires the models to have expert intelligence in order to modify text with a deep understanding of domain knowledge and structures. To bridge the gap between doctors and patients, we proposed a new approach of text style transfer for non-parallel data. Our proposed approach is based on masking expert terms and denoising autoencoder. We trained and tested our approach on MSD dataset and achieved a stable score across content similarity, perplexity, and style accuracy metrics
  • Thumbnail Image
    Item
    StructGAN: Image Restoration Maintaining Structural Consistency Using A Two-Step Generative Adversarial Network
    (Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur, Bangladesh, 2021-03-30) Zahin, Nahian Muhtasim; Rahman, Md. Mushfiqur; Mahmud, Kazi Raiyan
    Image restoration deals with the removal of noise, blurriness, missing patches, and other kinds of distortions in broken images. Traditional reconstruction and restoration approaches suffer from different kinds of limitations. In our work, we have improved upon those models by introducing novel structure loss that emphasizes the overall image structure rather than individual pixels. Our proposed model StructGAN can achieve a higher SSIM (Structural Similarity Index Measure) score while not massively compromising other noise metrics. Overall, our proposed model uses generative adversarial networks with a two-step generator network, a dual discriminator network, and coherent semantic attention (CSA) layer. The two-step generator helps refine the output. The dual discriminator ensures local and global correctness. The CSA layer ensures semantic consistency. Along with these, our model incorporates the novel structure loss. The structure loss is based on the Laplacian filter that calculates the overall structure-map of the image and tries to replicate the structure-map in the generation step. The results obtained by our model are qualitatively comparable to the performance of the state-of-the-art models. For certain metrics, e.g. SSIM, StructGAN quantitatively outperforms other models.
  • Thumbnail Image
    Item
    Object Tracking using End-to-End Detection and Deep Association Metric
    (2021-03-30) Yasmeen, Arowa; Rahman, Fariha Ishrat; Hassan, Inara Zahin
    Object Tracking has multiple major applications such as video surveillance for se- curity, traffic control, contact tracing, human computer interaction, gesture recog- nition, augmented reality, video editing, robotics etc. Often, to perform real-time tracking, video surveillance applications forgo detection accuracy in favour of de- tection speed. This paper proposes a combination of object detection and object tracking algorithms that gives an improvement on both detection accuracy and speed compared to existing video surveillance solutions. It also includes a method to trace the movement of a target from video surveillance footage and visualise the target’s path on a 2D map
  • Thumbnail Image
    Item
    Improving Hyperledger Fabric
    (Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur, Bangladesh, 2021-03-30) Haque, Ebtesam Al; Iffat, Fabiha; Aura, Novera Tasnuba
    Hyperledger Fabric is a popular open-source blockchain platform used by several projects around the world. However, several bottlenecks exist in the current system which limit its performance. One of the best attempts to improve Hyperledger Fabric was FasftFabric, which modifies the existing architecture to improve the throughput of Hyperledger Fabric.However, it does not provide an insight into the bottlenecks that could potentially exist in the orderer algorithm itself. In our work, we compared the existing ordering algorithms with Hashgraph to potentially improve Hyperledger Fabric’s throughput. Our proposed solution does not require a lot of system resources and also provides Byzantine Fault Tolerance with higher throughput.
  • Thumbnail Image
    Item
    Energy Efficiency of Mobile Edge Computing Systems
    (Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur, Bangladesh, 2021-03-30) Fariha, Raisa; Karim, Md. Ziad; Mahamud, Sheikh Faiyaz
    Quality of Service in case of IoT devices like Mobile Edge Computing Systems widely depends on the way the offloading of applications is done and how the devices are placed in the system. In case of Mobile Edge Computing Systems, the users are constantly moving so it is very important to find an optimal placement arrangement for them. But due to continuous movement, any optimal placement might even turn into an inefficient one within minutes. So, it is very important to design the placement of the applications keeping in mind the dynamics of the whole system. Again, the energy consumption done by the server in a MEC is an integral part while calculating the cost of services of the system. That is why in our paper we will address the problem of the optimal placement of the devices in a MEC as a multi-stage stochastic program. Our main goal will be to improve the Quality of Services as much as possible. Each of the mobile edge servers has a specific energy budget which we will also take into account. We have designed an algorithm to solve this problem and we will perform an experimental analysis to evaluate the performance of our designed algorithm.
  • Thumbnail Image
    Item
    Efficient Two-Stream Network for Violence Detection using Separable Convolutional LSTM
    (Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur, Bangladesh, 2021-03-30) Islam, Md. Zahidul; Rukonuzzaman, Mohammad; Ahmed, Raiyan
    Automatic detection of violence from surveillance footage holds special significance among the various subsets of general activity recognition tasks due to its broad applicability in autonomous security monitoring systems, web video censoring, etc. In this paper, we propose a two-stream deep learning architecture based on Separable Convolutional LSTM (SepConvLSTM) and pre-trained truncated MobileNet, in which one stream processes difference of adjacent frames and the other stream takes in background suppressed frames as inputs. Fast and efficient input pre-processing techniques were used to highlight moving objects in frames by suppressing nonmoving backgrounds and capturing motion in between frames. These inputs assist in producing discriminative features as violent activities are predominantly characterized by rapid movements. SepConvLSTM is built by replacing each ConvLSTM gate’s convolution operation with a depthwise separable convolution, resulting in robust long-range spatio-temporal features with significantly fewer parameters. We experimented with three fusion strategies to merge the output feature maps of the two streams. Three standard public datasets were used to assess the proposed methods. On the larger and more difficult RWF-2000 dataset, our model outperforms the previous best accuracy by more than 2%, while matching state-of-the-art results on the smaller datasets. Our studies demonstrate that the proposed models excel both in terms of computational efficiency and detection accuracy.
  • Thumbnail Image
    Item
    Bangla Sign Language Dataset Generation using Depth Information
    (Department of Computer Science and Engineering(CSE), Islamic University of Technology(IUT), Board Bazar, Gazipur, Bangladesh, 2021-03-30) Rayeed, S. M.; Akram, Gazi Wasif; Zilani, Golam Sadman
    Sign Language Recognition (SLR) targets on interpreting the sign language into text or speech, so as to facilitate the communication between deaf-mute people and ordinary people. This task has broad social impact, but is still very challenging due to the complexity and large variations in hand actions. Existing dataset for SLR in our country is based on RGB images (converted to grayscale) and CNN is used for classification. However, it is difficult to design model to adapt to the large variations of hand gestures in the dataset, also the computational expense is high. Modern researches on other sign languages have shown that using depth information in Sign Language Recognition (SLR) gives better accuracy, which hasn't been introduced yet in our country. In this paper, we intend to build a complete Bangla Sign Language (BSL) dataset using depth information from depth images. In order to do, we’ll be collecting our depth information from our captured image samples using MediaPipe which is a cross-platform framework for. building multimodal applied machine learning pipeline. It is quite a new and advanced technology in hand tracking and gesture recognition. As opposed to the existing image dataset, we intend to build a feature-based depth dataset, that, if accurately modeled, is more likely to give better results than the existing one in sign language recognition (SLR).