Thumbnail Image

Theses

Browse

2015 - 201936
Reset filters
Subject: Computer Engineering × End date: 2019 × Start date: 2015 × Type: Thesis ×
Reset filters

Search Results

Now showing 1 - 9 of 36
  • Thumbnail Image
    ThesisItemOpen Access
    Alternate approach of design of s-boxes and their evaluation
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-08) Pandey, Anukriti; Singh, B.K.
    Security of data is a major issue in performing any business or economic transactions. In order to provide effective security to important data communication, cryptography has a role to allow only the authorized person to access the data. There are many algorithms. Data Encryption Standard (DES) which is a symmetric block cipher algorithm with 64-bit blocks and a 56-bit key has been used for data encryption. Small key size makes it vulnerable to the Brute Force attack. At the very instance, it provides flexibility against the Differential Cryptanalysis and Linear Cryptanalysis. To enhance the security of DES, S-boxes are used to provide confusion (making relationship between key and cipher text more complex as possible) during the encryption process. In this thesis, alternate S-Boxes are formed using the XOR operation which is applied to the randomly selected values lies in the range 0-15. These S-Boxes are analyzed on the basis of Strict Avalanche Criterion (SAC) through the Avalanche Effect and Completeness Effect for which results are interpreted using the column graphs. Further, it is evaluated on an alternate of the function approximation process known as the Vandermonde Approach. It performs the comparison among functions of both the DES S-Boxes and alternate S-Boxes. The result shows that the proposed SBoxes satisfy the SAC criteria and provides better function approximation leads to the better data encryption.
  • Thumbnail Image
    ThesisItemOpen Access
    Deep learning approaches for detection of disease in potato and mango leaves
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-08) Arya, Sunayana; Singh, Rajeev
    Agriculture productivity has a big contribution to every nation's economy. Nowadays disease persistence in crops and plants are a major concern for farmers, so the detection process of diseases plays a very effective part in the farm sector and for farmers. To protect productivity, quantity, and quality of plants, proper care is mandatory. The traditional phenomenon requires a huge amount of work, time and continuous monitoring of farm for disease classification and detection. The latest methods in E-Agriculture for identification and detection of diseases like image processing, machine learning, and deep learning have been widely used. Deep learning uses Convolutional Neural Networks (CNN) for image classification as it gives the most accurate results in solving real- world problem. CNN has various pre-trained architecture like AlexNet, GoogleNet, DenseNet, SqueezeNet, ResNet, VGGNet etc. In this thesis, we have used CNN and AlexNet Architecture for detecting the disease in Mango and Potato leaf and compare the accuracy and efficiency between these architectures. The dataset containing 4004 images were used for this work. The images for potato were taken from plantvillage website (which is an open database of images published in 2016), while images for mango were collected from GBPUAT field location. The results show that accuracy achieved from AlexNet is higher than CNN architecture
  • Thumbnail Image
    ThesisItemOpen Access
    NetFlow based cyber threat classification using J48 and Random Forest machine learning algorithms
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-08) Rakesh Kumar; Singh, Rajeev
    In the field of Information Technology, cyber security plays a vital role. Securing information is the biggest challenge now a days. As the word cyber security comes in our mind the fear of cybercrime also comes in our mind at the same time. Cyber threats are nothing but an activity by which any targeted system can be compromised by altering the availability, integrity, and confidentiality of the system. To overcome such type of threats there are number of mechanisms available. Recently the Machine Learning (ML) approaches have proved to be a milestone for the detection of cyber threats using classification of NetFlows. The NetFlow is a network protocol designed by CISCO which is used to collect the network traffic (NetFlows). In this research work J48 and Random Forest (RF) machine learning algorithms are used for classification of cyber threats using NetFlows. The results are obtained by applying classification algorithms on NetFlows using Weka ML tool and RStudio. A comparison is made in various perspectives like accuracy, true positive (TP), false positive (FP), etc.
  • Thumbnail Image
    ThesisItemOpen Access
    Climate based factor analysis and epidemiology prediction for potato late blight using machine learning approaches
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-08) Tiwari, Pratima; Singh, B.K.
    Potato is one of the largest food crop and an integral part of world’s food supply. Late Bligh in Potato is community disease and has capability to devastate the entire crop rapidly.Estimated average annual loss form PLB is 15% around the world. Aim of the presented wok is to find a relationship between climate factors and late blight disease occurrence and severity. In presented framework of this thesis, the task of Factor Analysis and epidemiology prediction are assigned to different machine learning approaches. SVM approach is used to address Factor Analysis and ELM is used for Epidemiology prediction for Potato Late Blight Disease. Factor Analysis Model calculate the weights of the Climate based parameters depending on their relevance in deciding the blight and blight free year. Nine Climate based parameters are used in binary SVM model. The feature subset selected using SVM are used as input to ELM for epidemiology prediction along with the age of the plant. Two databases are prepared from AICRP and Climate Data, one for Factor Analysis with -1 and 1 binary class labels and one for Epidemiology prediction with five class labels (1-5). Database for Epidemiology Prediction is further divided into three sub databases for three separate planting dates. Analysis of the experimental results for Factor analysis shows that Maximum temperature, Maximum and Minimum Humidity, Sun Shine hours and Evaporation plays major role in occurrence of late blight disease. Experiments are conducted for Epidemiology Prediction and Comparison of the performance is demonstrated with other activation functions and different partitions of database to show its effectiveness and efficiency. On the basis of obtained results, SinC activation Function outperformed sigmoid function and has promising accuracy for all the data partitions.
  • Thumbnail Image
    ThesisItemOpen Access
    Extreme learning machine approach for prediction of forest fires using topographical and metrological data of Vietnam
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-08) Nikhilesh Kumar; Singh, B.K.
    A problem is a well-posed problem if it satisfy, solution existence, solution uniqueness and non- perturbation conditions. Ill-posed problems or inverse problems are the special case of well-posed problem, because inverse of major function does not exist. Forest fire prediction is an inverse problem. Forest are the largest natural resources. Forest fire is a calamity, it is a threat to the entire regime of flora and fauna. Forest fire mitigation is essential because it can devastate biodiversity, wild life and can cause economic loss. In the proposed work extreme learning machine is used, because it has capability prediction problem solves with better generalization and fast learning speed. ELM is a new approach to be used for forest fire prediction. Presented work predict the forest fire occurrence with the help of topographical and metrologicaldata, with parameters slope, Aspect, Elevation, NDVI, Distance to road, Distance to residential area, Land use, Temperature, Wind speed, Rainfall and forest fire occurrence. The motivation behind this work is to predict the forest fire to provide better way of management for this tragedy. In this research work a relationship is being established between forest fire causing factors and forest fire occurrence using historical data. The used database is already existing data of 540 historical locations of Vietnam. Experiments are conducted on different data partitions of availed data with different activation functions. On the basis of accuracy of model, sigmoid function found to be best and suggested to be used further for forest fire prediction.
  • Thumbnail Image
    ThesisItemOpen Access
    A fuzzy logic based model for detecting leaf blast disease in rice crop with ANFIS and linear regression technique
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-07) Chilwal, Bhavna; Mishra, P.K.
    This model uses the fuzzy logic advancement in detection analysis for agriculture sector using fuzzification tool in MATLAB. It is considered to be a good technique in precision agriculture. In this work, the first focus is to detect the level of leaf blast disease in rice crop by using three specific symptoms. The detected levels of disease grading level are Level 1- Mild, Level 2- Moderate, Level 3- Extreme corresponding to the disease occurrence. Then this disease grade level will be used to obtain the risk level in percentage form. The techniques used here are the fuzzy logic implementation in two level, the linear regression technique to identify the relationship between the disease level and risk level also dataset was generated by regression then used in ANFIS tool in MATALB and finally the decision tree is used to detect disease and the major symptom among all symptoms and which plays vital role in deciding the disease severity class label.
  • Thumbnail Image
    ThesisItemOpen Access
    Network traffic analysis based IoT botnet detection using Honeynet data, applying classification techniques
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-08) Banerjee, Mahesh; Samantaray, S.D.
    In recent years with the inception of Internet of Things (IoT) devices the internet is flooded with low powered connected devices. This has led to a scenario where attackers can launch catastrophic DDoS attacks using these IoT devices for disrupting the internet. In 2016, the a DDoS attack launched by Mirai botnet hits targets with a bandwidth of 620 Gbps and took down websites as such as Twitter, Netflix, GitHub, etc and almost stalled the internet on the U.S east coast. Developing effective and sufficient approach for controlling IoT Botnet attacks is still a challenge. In this thesis a Network Traffic analysis based IoT botnet detection techniques has been developed.The detection of botnets is done by deploying a honeynet which provides us with activity logs of the intrusion attempts as well as the network traffic dump in the form of packet capture. The network traffic is used for extracting the flow of the traffic. Our work focuses on botnet detection using the network flows and applying classification techniques for finding features which have significant clue for botnet attack. For implementation, local honeynets namely CDAC CTMS and PantHoneynet have been used which are deployed at Department of Computer Engg. GBPUAT, Pantnagar. The data set obtained from the honeynets are used for the detection of botnets by the means of supervised machine learning classification techniques such as Random Forest ensemble classifier its performance with various other supervised machine learning algorithms is also compared.
  • Thumbnail Image
    ThesisItemOpen Access
    An integrated approach for cyber attack prediction using Honeynet and Socialnet data, applying improved association rule mining technique
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2019-08) Agarwal, Bhavna; Samantaray, S.D.
    Cyber attacks are becoming more lethal in last couple of years, as it has become a parallel industry for targeting Government, Defence, Nuclear and other High-profile organisations. People usually commit cyber-crimes for demanding ransom, for fame or due to personal grudges. Lombroso theory for crime psychology uses certain traits of a person to identify whether that person is a criminal. Similarly understanding the Crime psychology of attackers help in mitigating the cyber attacks. Finding associations of Crime with spatial and temporal parameters such as locations, time, event and activity is still an open research area. In the thesis an Integrated approach for cyber attack prediction is proposed on the basis of the data acquired from the Honeynet and Socialnet. For implementation, local Honeynets namely CDAC CTMS and Pant Honeynet have been used which are deployed in Department of Computer Engineering at GBPUAT, Pantnagar. The Socialnet data is acquired using News Tracker API. The Honeynet provides us attack data in the form of activity logs while the Socialnet provides us events which correlate with the cyber attacks. The attack data and events obtained from Honeynet and Socialnet respectively are used to form an Augmented Transaction Database. Then Improved Association Rule Mining Technique is applied for producing Association Rules. These Association Rules help in predicting the occurrence of cyber-attack which would assist in providing better advisory and deployment of security measures.
  • Thumbnail Image
    ThesisItemOpen Access
    Image classification using convolutional neural network with tensorflow
    (G.B. Pant University of Agriculture and Technology, Pantnagar - 263145 (Uttarakhand), 2018-08) Gangwar, Krishanveer; Negi, C.S.
    The present work proposes a methodology for classifying images of cats and dogs accurately, using Convolutional Neural Network and Back-Propagation. The image recognition market is estimated to expand from US$15.95 Billion in 2016 to US$38.92 Billion by 2021, at the CAGR of 19.5% between 2016 and 2021. Facebook is the largest image sharing site on Internet. Images represent the largest source of data usage on Facebook. On an average, more than 300 million images are uploaded to its site daily. The present work is aimed to develop a model for classification of cats and dogs. The work initiate with image acquisition. Then, applying image pre-processing to bring all the images in required shape and format. Proposed work is divided into two parts features learning and classification. Feature learning is done through convolutional layer and pooling layers and classification is done by Fully connected layers. The training dataset contains 12108 images. Python 3.6 is used for the programming and system is tested for 1920 images and gained the accuracy of 85.36%. The system is also validated using holdout validation techniques.