Thumbnail Image

Theses

Browse

2013 - 201926
Reset filters
Subject: Information Technology × End date: 2019 × Start date: 2010 ×
Reset filters

Search Results

Now showing 1 - 9 of 26
  • Thumbnail Image
    ThesisItemOpen Access
    A novel approach to improve the Z-SEP protocol in WSNs
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-08) Dhek, Ankita; Mandoria, H.L.
    The eminence of Wireless Sensor Networks (WSNs) has grown massively in the past years because WSN has the possibility to associate the physical world with the virtual world by framing a system of various sensor nodes by deploying it in a network. Here, sensor nodes are typically battery-driven gadgets, and consequently energy consumption of sensor nodes is a noteworthy design issue. To handle this issue Cluster based routing is most popular routing technique in Wireless Sensor Networks(WSNs). Due to various need of WSN applications, efficient energy utilization in routing protocols is still a potential area of research. The proposed work tried to overcome the problem of energy contraint in Z-SEP protocol in such a way that energy consumption of every node in the wireless sensor networks is minimized with the help of improving the stability period, instability period ,network lifetime and throughput. This thesis solves the problem by proposing a clustering based single-sink routing protocol for heterogeneous wireless sensor networks in which 3 types of nodes are introduced. The idea of multi-hop communication and distance factor is considered.Using the MATLAB simulation the result is compared with the Z-SEP protocol and the result shows that the proposed protocol outperforms the Z-SEP protocol.
  • Thumbnail Image
    ThesisItemOpen Access
    Intrusion detection using ensemble methods and deep learning
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-08) Rai, Mahima; Mandoria, H.L.
    In recent years Machine Learning and Deep Learning have played a crucial role in the field of cyber security. This research therefore intends to design IDS with a good mix of various ML and DL classifiers. It gives us an idea about which classifier gives the highest accuracy. The aim of this research is to classify whether the packet is a normal data packet or attacked data packet. Therefore the model of IDS should be implemented based on effective and significant ML and DL algorithms. To ensure network security, an effective intrusion detection system is required. Several ensemble methods like XG-Boost and LGBM have been developed in the past 4-5 years. These have not been exploited in the previous researches on anomaly detection. Our study will make use of these novel Gradient Boosting Decision Tree algorithms. XG-Boost and LGBM have proved to be the most productive techniques for several supervised and unsupervised learning tasks. In this research, we study several machine learning and deep learning classifiers and compare their performance. We have used the NSL KDD dataset to predict the probability of occurrence of 21 different classes of attacks on a network and three different categories of models - Linear Models including Logistic Regression and Stochastic Gradient Descent (SGD) classifier; Gradient Boosting Decision Tree ensembles including LightGBM (LGBM) and XG-Boost; and a Deep Neural Network (DNN) classifier and also trained a stacked model consisting of all these models as base learners. Finally, compared the performances of all the models for Network Intrusion Detection using the NSL-KDD dataset and have drawn useful conclusions. The simulation result shows that Light GBM, XGBoost, and stacked classifiers outperform with high accuracy as compared to Logistic regression, Stochastic Gradient Descent Classifier and Deep Neural network. All of these are predictive analysis techniques.
  • Thumbnail Image
    ThesisItemOpen Access
    Analysis of Satellite Images using Support Vector Machine (SVM)
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2018-08) Arya, Diksha; Singh, Rajesh
    Satellite captured images are eyes in the sky that contain information from around the globe. The information derived from the Satellite Images is helpful in Remote Sensing Applications, Research and Analysis, Organizations or Government bodies in monitoring space, Civil defence operations, etc. Satellites captures huge collection of images at a regular interval of time and analyzing those images manually is very difficult and time consuming. Therefore, an ideal picture classifier system is required to be located that intend to classify the images captured from the satellites so that the images of interest can be easily retrieved. Thus, in our research we aimed to propose a Satellite Image Classification system that can automatically classify the category of physical scene present in an image. For achieving our objective we have utilized Support Vector Machine (SVM), a supervised machine learning algorithm for performing classification; two widely used techniques for extracting Features which are Grey Level Co-occurrence Matrix (GLCM) and Gabor Filter; and Fuzzy C Means for Image Segmentation. We have performed Satellite Image Classification for five physical categories namely Desert, Mountain, Residential, River and Forest. We measured our classification system accuracy using confusion metrics and calculated the precision, sensitivity, specificity and F1 Score. Our classification system achieved overall accuracy of 91.66%.
  • Thumbnail Image
    ThesisItemOpen Access
    A novel chaotic based keyless encryption technique for secured transmission of images
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2018-08) Singh, Chetana; Pandey, Binay Kumar
    These days secured data storage and transmission have become an important issue in this digital world due to increased use of internet for communication purpose. Information security is becoming more important as the amount of sensitivity data being exchanged on the Internet increases. The service like confidentiality and data integrity are required to protect data against unauthorized usage and modification. In these years several image encryption methods are introduced by various researches to secure multimedia information while transmit via public networks. A novel Keyless Image encryption method based on chaotic map is suggested in this research work. Each component of image is encrypted by shuffling pixels and this shuffling is decided by the modified cat map. To transmit data whole image is convert into linear form then modulate the linear data using modulation technique OQPSK. The results confirm that the proposed method resist the statistical analysis. Also attains acceptable correlation coefficient value and has a robust performance against attacks. The simulation of the above algorithm is done on Matlab.
  • Thumbnail Image
    ThesisItemOpen Access
    Fuzzy based semantic clustering of news articles
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2018-10) Priyanka; Joshi, Sanjay
    Text mining is a process that uses data mining approaches to extract valuable information held in the hidden form in textual data. In this paper, a framework for fuzzy clustering of news articles is proposed. These news articles originate on different news portals on the web. The data sets are fetched from two different Indian news portals, The Hindu archive and Times Of India archive. Six data sets are used for implementation and evaluation: 4 news articles Times of India, 150 news articles Times of India, 1000 news articles Times of India, 4 news articles The Hindu, 150 news articles The Hindu, 1000 news articles The Hindu. The fetched data is stored in a central database and then preprocessing reduces the noise. Tokenization is done to split the text content into separate words. Stop words are removed from the text data as they have no significance for cluster discrimination. Then lemmatization technique is applied. Tf-idf is calculated for the data set and saved in the word frequency vector. On these vectors, distance measure or similarity measure function is used to find the similarity between articles. Tf-idf with cosine similarity measure gives semantic similarity between articles. One article may belong to more than one cluster so fuzzy membership values must be generated. The articles are clustered using two clustering algorithms k-means clustering and fuzzy c-means clustering. The similar documents are grouped into same cluster and dissimilar documents are put into different clusters. The proposed framework shows that fuzzy clustering does not restrict each news article to belong exactly to one cluster. Therefore this framework when applied to information retrieval systems or other application systems, system gives better performance and relevance to the users.
  • Thumbnail Image
    ThesisItemOpen Access
    Performance evaluation off content based image retrieval using dwt, modified k means and neural network classification
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2018-10) Aeri, Manisha; Ashok Kumar
    The increasing need for searching relevant images in large scale databases is an active area of interest now days. The current method of text based image retrieval has many challenges as context sensitive images cannot be retrieved, the effort required to annotate each and every image manually as well as the difference in human perception while describing an image gives inaccurate and inefficient results during the retrieval process. So, Content Based Image retrieval came into picture in which the retrieval of images can be done by using the contents of an image such as color, shape or texture. In this research work we have proposed a novel approach for content based image retrieval by applying an efficient clustering algorithm i.e. Modified k means for image segmentation and we have used the concept of discrete wavelet transformation, color moments and HSV histogram for extracting image features. Artificial neural network is trained about the extracted features from the database images. The testing phase involves the querying and retrieval task in which the query images features are compared with the trained features of database images and the best matched images are retrieved from the database similar to the query image. This technique is tested by conducting experiments on WANG image dataset containing 1000 general purpose colored images in terms of precision and recall in which the proposed technique gave better results as compared to the existing techniques.
  • Thumbnail Image
    ThesisItemOpen Access
    A novel approach recognizing objects from images by using SIFT and HMM
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2018-08) Verma, Aanchal; Srivastava, Ratnesh Prasad
    Recognizing Objects in images and finding a particular image from the set of images have always been a challenging task. There has been significant progress in this area as discussed in literature survey part of the thesis. This section proposed a novel approach for recognizing such objects by the use of algorithms SIFT and HMM which are used for the purpose of feature extraction and classification. The survey papers of last few years clearly shows that there has been a missing in the use of SIFT and HMM for recognizing the images. Therefore, The proposed framework will perform the recognition of an object as human or something by using database images and store the derived features or Keypoints from image sequence for measurement in recognition stage. An automated extraction of relevant feature point from given image is provided to automate the recognition procedure using SIFT. This process improves the recognition accuracy. The human facial features extracted through SIFT are utilized for the recognition of human. The SIFT feature will be created for each given images and the key points are computed, and then HMM is applied for recognition. HMM uses SIFT feature to perform recognition process on MATLAB. The recognition outcomes demonstrate that our proposed framework gives more accurate performance when comparing to tradition procedure.
  • Thumbnail Image
    ThesisItemOpen Access
    An energy efficient mechanism for data collection in wireless sensor network
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2013-07) Tayal, Pranavi; Mandoria, H.L.
    A sensor network is composed of vast number of tiny sensor in a limited area. Each sensor is defined with some energy parameters and the energy based constraints. According to these constraints, as the communication is performed, each participating node loses some amount of energy. Multicast and Broadcast communication are the basic communication requirements of a sensor network. But such kind of communication increases the network traffic extensively as well as gives large amount of energy loss. Data collection provides the solution to this problem by combining the multiple communications in single communication path. Data collection is one of the major communication approach in which multiple sources are sending data to single sink. In this present work, an agent based approach is defined to generate the effective aggregative path so that the network life and communication will be improved. The presented approach is divided in two main stages and both stages are controlled by multiple agents distributed over the network. In the very first stage, the agent will perform the analysis over the network and assign the weightage to each node based under the different parameters. Once the weights are assigned to each node the next work of agent is to generate the aggregative path. This path generation is based on multiple parameters. The parameters considered in this work to generate the effective path are loss rate, response time and the communication delay. The presented research work is about to generate an effective communication path so that the effective communication will be performed. The path generation process is divided in two main stages, first phase is to identify frequency (load) of each node and second to generate path so that load balancing will be improved. Most frequent node here represents the heavy load node over the network. The next work is to calculate the path from source to destination by comparing loads of neighboring nodes and here the current node select that neighbor node which is having low load among list of neighboring nodes and then perform the communication over that node. The presented work is implemented in NS2 environment and obtained results shows that the presented work has improved the network throughput extensively and reduced the network delay and the data loss.
  • Thumbnail Image
    ThesisItemOpen Access
    An additive noise suppression in noisy signal using spectral subtraction
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2017-07) Malik, Medha; Negi, Chetan Singh