Skip to main content
main-content

Über dieses Buch

This book gathers a collection of high-quality peer-reviewed research papers presented at First International Conference on Innovations in Software Architecture and Computational Systems (ISACS 2021), held at Guru Nanak Institute of Technology, Kolkata, India, during 2 – 3 April 2021. The book primarily focuses on developing artificial intelligence-based algorithms and methodologies for enabling intelligent hardware and software systems. This book brings together the latest findings on efficient technological solutions for developing intelligent and hybrid systems, intelligent software architecture, machine intelligence-based analytical tools and also smart sensors and networks. The prime focus is on solving technological problems using state-of-the-art research finding like fuzzy computing, evolutionary and hybrid frameworks, neuro-computing, etc., along with other AI-based computation platforms. The book offers a valuable resource for all undergraduate, postgraduate students and researchers interested in exploring solution frameworks for social good problems using artificial intelligence.

Inhaltsverzeichnis

Frontmatter

Chapter 1. Text-to-Image Classification Using AttnGAN with DenseNet Architecture

Abstract
In this attentional generative adversarial networks (AttnGAN) for text-to-image conversion, we have used a CUB dataset with 12,000 images of 200 different birds with 10 captions for each image (12,000 * 10 captions). In random distribution splitting, we divided the data into training and testing sets with 49.2% of data and 50.8% of data, respectively. This method is able to synthesize fine-detailed images by the use of a global attention that gives more attention to the words in the textual descriptions. Also we have the deep attentional multimodal similarity model (DAMSM) that calculates the matching loss in the generator. Though this work produced images of high quality, there was some loss while training the system and it takes enough time for training. This paper proposes the DenseNet architecture with AttnGAN in order to reduce the loss and training time thereby synthesizing images with more distinct features. This technique was able to reduce the loss by 1.62% and could retrieve faster results by 768 s per iteration than the existing CNN architecture.
Anunshiya Pascal Cruz, Jitendra Jaiswal

2. Cyclone Detection and Forecasting Using Deep Neural Networks Through Satellite Data

Abstract
Satellite imagery provides the initial data information in cyclone detection and forecasting. To mitigate the damages caused by cyclones, we have trained data augmentation and interpolation techniques for enhancing the time-related resolution and diversification of characters in a specific dataset. The algorithm needs classical techniques during pre-processing steps. Using 14 distinct constraint optimization techniques on three optical flow methods estimations are tested here internally. A Convolutional Neural Network learning model is trained and tested within artificially intensified and classified storm data for cyclone identification and locating the cyclone vortex giving minimum of 90% accuracy. The work analyzes two remote sensing data consisting of merged precipitation data from TRMM and QuikSCAT wind satellite data and other satellites for feature extraction. The result and analysis show that the methodology met the objectives of the project.
Shweta Kumawat, Jitendra Jaiswal

3. An Improved Differential Evolution Scheme for Multilevel Image Thresholding Aided with Fuzzy Entropy

Abstract
Image segmentation problem has been solved by entropy-based thresholding approaches since decades. Among different entropy-based techniques, fuzzy entropy (FE) got more attention for segmenting color images. Unlike grayscale images, color images contain 3-D histogram instead of 1-D histogram. As traditional fuzzy technique generates high time complexity to find multiple thresholds, so recursive approach is preferred. Further optimization algorithm can be embedded with it to reduce the complexity at a lower range. An updated robust nature-inspired evolutionary algorithm has been proposed here, named improved differential evolution (IDE) which is applied to generate the near-optimal thresholding parameters. Performance of IDE has been investigated through comparison with some popular global evolutionary algorithms like conventional DE, beta differential evolution (BDE), cuckoo search (CS), and particle swarm optimization (PSO). Proposed approach is applied on standard color image dataset known as Berkley Segmentation Dataset (BSDS300), and the outcomes suggest best near-optimal fuzzy thresholds with speedy convergence. The quantitative measurements of the technique have been evaluated by objective function’s values and standard deviation, whereas qualitative measures are carried out with popular three metrics, namely peak signal-to-noise ratio (PSNR), structural similarity index measurement (SSIM), and feature similarity index measurement (FSIM), to show efficacy of the algorithm over existing approaches.
Rupak Chakraborty, Sourish Mitra, Rafiqul Islam, Nirupam Saha, Bidyutmala Saha

Chapter 4. Clustered Fault Repairing Architecture for 3D ICs Using Redundant TSV

Abstract
Through silicon via (TSVs) base 3D integrated circuit (3D IC) has become most emerging technology in semiconductor industry. TSVs-based 3D-IC has some advantages like low foot-print area, less power requirement, small interconnection length, etc. The major concern is manufacturing defect in for 3-DIC. There may have some manufacturing defect in TSVs. A single fault can destroy the total chip. So, it is required to repair faulty TSV to make a chip functional. An effective repairing method is required. The one solution would be use of redundant TSVs the reroute the signal. As of now, there are very few works exists to address problem of clustered fault for irregularly distributed TSVs all over the chip. Some works exist on regularly distributed TSVs but very few in irregularly distributed TSVs. In our work, we have proposed a method to form groups among functional and redundant TSVs and make connection between functional and redundant TSVs in an innovative way that we can repair clustered fault. Also, we tried to use minimum number multiplexer (MUXs) so that area overhead will be reduced.
Sudeep Ghosh, Mandira Banik, Moumita Das, Tridib Chakraborty, Chowdhury Md. Mizan, Arkajyoti Chakraborty

Chapter 5. Study on Similarity Measures in Group Decision-Making Based on Signless Laplacian Energy of an Intuitionistic Fuzzy Graph

Abstract
In sight of intuitionistic fuzzy inclination relations (IFIR), we study group decision-making (GDM) problems. We propose another way to deal with assess the relative notoriety weights of specialists by registering the questionable proof of intuitionistic fuzzy inclination relations and the normal similitude level of one individual intuitionistic inclination connection to the others. This new approach takes both objective and subjective evidence of specialists into consideration. Then, we assimilate the weights of authorities into the precise intuitionistic fuzzy inclination relations and progress a relative similarity method to originate the significances of substitutes and better of the substitutes. The balance investigation with extra techniques by two numerical examples shows the sober mindedness and supportiveness of the anticipated strategies.
Obbu Ramesh, S. Sharief Basha, Raja Das

Chapter 6. Asymptotic Stability of Neural Network System with Stochastic Perturbations

Abstract
In this article, we have considered a mathematical model of a neural network. The model is characterised by a delay difference equation with stochastic perturbations. We have proved the condition of asymptotic stability behaviour of the trivial solution of s multiple-delay model with stochastic terms.
M. Lakshmi, Raja Das

7. Uniform Grid Formation by Asynchronous Fat Robots

Abstract
In this paper, we introduce a distributed algorithm for the formation of a grid pattern by multiple homogeneous, autonomous, disc-shaped robots (also referred to as fat robots). The robots organize themselves to form a grid (not given in prior) which is required for covering, safeguarding or supervising a geographical region by the robots. The operation of the robots is based on execution cycles of the phases ‘wait–look–compute–move’. In the ‘wait’ phase, the robots are inactive or dormant; in the ‘look’ phase, they observe within their visibility range; in the ‘compute’ phase, the robots decide a destination to move to; and finally, they advance to their decided target point in the ‘move’ phase. Eventually, the robots distribute themselves to create a grid. They are oblivious, i.e. the robots cannot recollect any past action or looked data from a previous cycle, and there is no interaction among these robots by message passing. They are anonymous, identical and have unlimited visibility. Though the robots are see-through or transparent, they are physical obstruction for the others. The algorithm proposed here also ensures the prevention of deadlock and collision among the robots.
Moumita Mondal, Sruti Gan Chaudhuri, Punyasha Chatterjee

8. A LSB Substitution-Based Steganography Technique Using DNA Computing for Colour Images

Abstract
Steganography is the process of using a cover or medium such as photograph, audio, text and video to shield information from the outer world. This paper suggested a new approach based on DNA computing for hiding information within an image using the least significant bit (LSB). To do this, the DNA is decomposed by four nucleotides, namely adenine, thymine, guanine and cytosine. Whereas a codon is a sequence of three nucleotides and to represent a codon, the two-pixel LSBs from the image are taken and then converted into protein. The confidential data bits are concealed into the codons, which transmutes the original cover image into a stego-image which is completely trustworthy to avoid human visual system, and the confidential data is impossible to detect. The empirical findings show the effectiveness of the suggested approach by producing 0.750 bpp of hiding power with average 56.48 dB of peak signal-to-noise ratio (PSNR) which makes it a strong image steganography technique.
Subhadip Mukherjee, Sunita Sarkar, Somnath Mukhopadhyay

Chapter 9. An Approach of Safe Stock Prediction Using Genetic Algorithm

Abstract
Stock market investments are an admired problem but onerous task. Because it is very unstable in nature for different factors and it is very hard to predict the safe stocks to invest at different circumstances. To guess the safe stocks to invest from a very large no of shares is an attractive research area that needs to be done efficiently because it is the question of profit and loss. In this research work, an optimized search algorithm has been used to predict no of the stocks from a huge scale of shares or stocks in which it will be safe to invest. In this proposed work, NIFTY top 50 shares to which it will be safe to invest for long-term investment has been taken into consideration and has been evaluated the safe stocks in a rank wise manner using an efficient search optimization technique, genetic algorithm (GA). Genetic algorithm is capable to yield better accuracy than other similar models. This is a heuristic search optimization method for searching of a very vast domain of dataset. This algorithm reflects the process of natural selection where the fittest individuals are selected for reproduction in order to produce offspring of the next generation. An insignificant population of individual archetypes can successfully search a large space because they comprehend schemes. Beneficial sub-structures that can be theoretically united to make fittest entities. Fitness is determined by investigating a huge number of distinct fitness cases. This procedure can be very effective if the fitness cases also grow by their individual GAs.
Nilanjana Adhikari, Mahamuda Sultana, Suman Bhattacharya

Chapter 10. Urban Growth Prediction of Kolkata City Using SLEUTH Model

Abstract
Due to increasing urbanization, urban growth or sprawl monitoring and measurement is needed in the developing countries like in India. For findings the urban growth and prediction of upcoming possible development of Kolkata city the SLEUTH model is employed. It is one of the most important urban development models which is utilized all over the world. The model was calibrated through the historical past data which are extracted from the satellite images in different time period. Six input data are used in this model such as slope, land use, excluded, urban extent, transportation, and hillshade. All the inputs were derived from the satellite image, and that was classified through the Maximum Likelihood classification technique. In this research, five Landsats temporal data (1978, 1988, 2000, 2010 and 2020) were used for prediction of urban growth of Kolkata City. The historical urban scenario is presented in this study which allowed urban expansion persistence in the previous trend. Calibration results show that the spread coefficient value is high which indicates the future prediction of Kolkata city is edge enlargement. Study revealed that in future more urban expansion may happen from 2020 to 2040 in the north-east and south-east positions of the Kolkata City. Besides, it is also observed that in future in the year of 2040 about 70% of total study area may be occupied by the urban area.
Krishan Kundu, Prasun Halder, Jyotsna Kumar Mandal

11. Suicide Ideation Detection in Online Social Networks: A Comparative Review

Abstract
Online social network has turned out to have widespread existence on the Internet gradually. Social network services allow its users to stay connected globally, help the content makers to grow their business, etc. However, it also causes some possible risks to susceptible users of these media, for instance, the rapid increase of suicidal ideation in the online social networks. It has been found that many at-risk users use social media to express their feelings before taking more drastic step. Hence, timely identification and detection are considered to be the most efficient approach for suicidal ideation prevention and subsequently suicidal attempts. In this paper, a summarized view of different approaches such as machine learning or deep learning approaches, used to detect suicidal ideation through online social network data for automated detection, is presented. Also, the type of features used and the feature extraction methods for suicidal ideation detection are discussed in this paper. A comparative study of the different approaches to detect suicidal ideation is provided along with the shortcomings of the current works, and future research direction in this area is discussed in this paper.
Sayani Chandra, Sangeeta Bhattacharya, Avali Banerjee(Ghosh), Srabani Kundu

Chapter 12. An Improved K-Means Algorithm for Effective Medical Image Segmentation

Abstract
Clustering-based image segmentation got wide attention for decades. Among various existing clustering techniques, K-means algorithm gained popularity for its better outcome. But the drawback of this algorithm can be found, when it is applied to noisy medical images. So, modification of the standard K-means algorithm is highly desired. This paper proposes an improved version of K-means algorithm called as (IKM) to get more effective and efficient outcomes. The efficiency of the algorithm depends on the speed of forming the clusters. So, in the proposed approach, new idea has been applied to find the minimum distance to generate the clusters. The proposed IKM algorithm has been applied to the set of noisy medical images, and the segmented outcomes have been evaluated by the standard quality measurement metrics, namely Peak-Signal-to-Noise-Ratio (PSNR) and structural similarity index measurement (SSIM). The outcomes have also been compared with the Watershed algorithm for showing the betterment of the proposed approach.
Amlan Dutta, Abhijit Pal, Mriganka Bhadra, Md Akram Khan, Rupak Chakraborty

Chapter 13. Breast Cancer Histopathological Image Classification Using Convolutional Neural Networks

Abstract
Nowadays, the classification of medical images has become an essential part of identifying the disease. Among various existing critical diseases, identification of breast cancer has now come up with the topic of investigation. To identify the affected regions of the images, a deep learning-based approach has got wide attention for decades. Convolutional neural networks (CNN) among all deep learning techniques proved their best efficiency in this field. In this paper, one improved CNN-based approach has been proposed to classify the breast cancer images obtainable from the standard PatchCamelyon (PCam) benchmark dataset. It is available for free from the website link https://​www.​kaggle.​com/​c/​histopathologic-cancer-detection/​data. In the improved model, various existing layers like convolutional, ReLU, pooling, fully connected have been added as well as modified for better efficiency and efficacy of the algorithm. Further to be added that Adam optimizer has been used here with cross entropy as a loss function. This improved model has been compared with two recent CNN-based approaches applied to medical datasets. The comparative outcomes suggest strong improvements in terms of classification accuracy (probably 2–3%) and computational time of validation loss for both training and testing data over existing models.
Ankita Adhikari, Ashesh Roy Choudhuri, Debanjana Ghosh, Neela Chattopadhyay, Rupak Chakraborty

Chapter 14. A Framework for Predicting Placement of a Graduate Using Machine Learning Techniques

Abstract
Campus placement carries a great significance for all the students and educational institutes. Nowadays, students give special attention to past placement records while selecting an institution for their admission. Hence, the institutions attempt to improve their graduate job appointment activities. The aim of this work is to evaluate past students’ academic records, and forecast placement probability of existing students. This placement predictor model takes different parameters those can be used to analyze the skill level of the student. While some parameters are taken from the institute level data, others are obtained from tests records conducted by the placement itself. Combining these data points, the model predicts, whether a student will be placed or not. Therefore, a model has been proposed to predict the placement possibilities with the help of machine learning algorithms. A framework is designed with the help of eight machine learning algorithms over the collected dataset, and the accuracy of the model is checked through these algorithms. Besides, random forest algorithm gives better performance among all the eight machine learning techniques.
Amrut Ranjan Jena, Subhajit Pati, Snehashis Chakraborty, Soumik Sarkar, Subarna Guin, Sourav Mallick, Santanu Kumar Sen

Chapter 15. A Shallow Approach to Gradient Boosting (XGBoosts) for Prediction of the Box Office Revenue of a Movie

Abstract
In the recent past, machine learning paradigms like the ensemble approaches have been used effectively to predict revenue from large volumes of sales data that helped the decision-making process in many businesses. The proposed work in this paper proposes a modified approach of ensemble algorithms to predict box office revenues of upcoming movies. A shallow version of the gradient boosting (XGBoosts) has been proposed to predict the box office revenue of movies based on several primary and derived features related to the movies in particular. Further studies have found that features such as budget, runtime, budget year ratio can also be considered as some of the more important estimators of the box office revenue. These features along with some other features have been used as an input to the proposed model in this proposed work to make significantly good predictions about the box office collection of a movie. The results are reported by testing and forecasting based on simulation on a standard data set. The precision of the model is tested using popular metrics such as R2, MSLE. The results reported gives efficacy of the proposed approach that can be further used in other business models words.
Sujan Dutta, Kousik Dasgupta

16. A Deep Learning Framework to Forecast Stock Trends Based on Black Swan Events

Abstract
The stock trends prediction is the key interest area for the investors. If the successful stock trends prediction is achieved, then the investors can adopt a more appropriate trading strategy, and that can significantly reduce the risk of investment. But it is hard to predict the stock market due to the unpredictable fatal events called Black Swan events. In this work, we propose a deep learning framework to predict the daily stock market trends with the intent that our framework can predict the stock market even on the time periods of the Black Swan events. In this framework, the signals of various technical indicators along with the daily closing price of the stock market and other influencing stock markets are used as the input for more accurate predictions. The base module of this framework is 1D convolutional neural network (1D-CNN) and bidirectional gated recurrent unit (Bi-GRU) neural network. We conduct vast experiments on the real-world datasets from two different stock markets and show that our framework exhibit satisfactory prediction accuracy for the normal circumstances. It outperforms other existing similar works during the periods of Black Swan events.
Samit Bhanja, Abhishek Das

Chapter 17. Analysis of Structure and Plots of Characters from Plays and Novels to Create Novel and Plot Data Bank (NPDB)

Abstract
This paper consists of a technique that helps to build a well-structured character interaction network from plays and novels. In addition, the resulting network describes the leading characters along with their gender. Moreover, the network analyzes the gender of the central characters to the plots preferred by the author. We propose a databank called Novel and Plot Data Bank (NPDB) to store the relevant information by computing informative properties of the resulting network.
Jyotsna Kumar Mandal, Sumit Kumar Halder, Ajay Kumar

Backmatter

Weitere Informationen

Premium Partner