8333 research outputs found
Sort by
Enhancing Small Object Detection in Aerial Imagery: A Comparative Study of YOLO and RT-DETR Models Using Slicing Aided Hyper Inference
No full textDetecting small objects in many aerial images caught by drones is now an important challenge that many experts face in the field of computer vision. Traditional object detection algorithms often struggle to identify small objects. Their low resolution and complicated aerial environments cause this issue. Traffic monitoring, urban planning, and disaster management will grow in relevance with this issue. The performance of deep learning models like YOLO and RT-DETR are investigated in this study. The Slicing Helped Hyper Inference technique is used to enhance these models for detection improvements. In this study, three models were trained on a VisDrone dataset for 200 epochs and their performance were compared with standard inference and SAHI techniques to determine their efficiency. The outcome of this study showed that with SAHI models were performing better, with higher mAP metrics, especially in detecting small objects. By increasing accuracy and reliability, many practical applications can greatly benefit from improved object detection for drone surveillance systems, which are important in many fields such as urban planning and emergency responses. SAHI technique and pre-trained models are trained and combined in this study, drawing attention to opportunities to solve problems in computer vision
Implementation of Advance Encryption Techniques to Protect Sensitive Financial Data from Cyber Threats
No full textThis research mainly focus to analyse and examine the effectiveness of advance encryption techniques (mainly cryptographic) in protecting the sensitive financial data from cyber threats within accounting and banking practices. The research covers the assessment of encryption methods / techniques adequacy and understand the subsequent impact on different aspects of sensitive financial data and related systems. The encryption techniques mostly covered cryptographic methods in this study. A two-pronged analytical methodology is adopted in the research i.e., Structural Equation Modelling (SEM) and matrix analysis. This research provides a technical aspect of cryptographic encryption techniques using matrix analysis concentrating on substitution ciphers including Hill cipher. The study outcome confirm the reliability of these techniques in successfully securing confidentiality and integrity of the sensitive financial data. The SEM analysis also identifies additional important results: first, the use of cryptographic techniques increases perceived level of security namely by 0.23 standardized path coefficients (β = 0.23, p = 0.032); second, regulatory compliance is positively impacted by variables like encryption that have a similar effect size (β = 0.48, p < 0.01). Additionally, an impact on satisfaction of end user is positively observed (β = 0.27, p = 0.03), though challenges persist in implementation, particularly regarding workforce technology infrastructure and training. The study underscores the critical role of strategic implementation in maximizing the efficacy of cryptographic solutions, emphasizing the need for user-centred design and alignment with evolving regulatory frameworks. It highlights the necessity of a balanced cybersecurity approach that integrates advanced encryption techniques with practical considerations for deployment. The research provides valuable insights both qualitative and quantitative to cyber and Information Technology (IT) practitioners, experts, standards and policymakers in establishing comprehensive data protection guidelines and processes in the financial sector
IOT Forensics: A Comprehensive Analysis of an IOT Device using Digital Forensic and Penetration Testing Tools and Methodologies
No full textThe domain of Internet of Things has been continuously expanding due to its increasing demand and diverse applications across Smart Homes, Healthcare, Virtual Assistance, and Agriculture. However, the security of these IOT devices is often neglected. In addition, the interconnected nature of these devices across unknown and heterogenous networks leads to an unprecedented number of IoT endpoints. This nature of complexity and security neglection gives rise to several cyber-attacks resulting in huge losses. This research focuses on executing a comprehensive analysis of a Baby Monitoring IOT device using Digital Forensics Tools and Methodologies. In this scenario, the sub-domain of Digital forensics i.e. IOT Forensics will be utilised to carry out a comprehensive analysis of a Baby Monitor highlighting its vulnerabilities, behaviour, and operational lifecycle. This research utilizes the NIST framework for digital forensics wherein the phases of Identification, Preservation, Analysis and Documentation will be implemented. In addition, the study also focuses on executing Vulnerability Assessment and Penetration Testing to identify and explore the types of threats and vulnerabilities pertaining to the Baby Monitor and provided mitigation and remediation solutions to minimise the impact for these vulnerabilities
Advanced ML Approaches for Intrusion Detection: A Comprehensive Analysis Using UNSW-NB15 and NSL-KDD Datasets
No full textThe exponentially of complex connecting systems in the information age has brought new and more difficult challenges in the cyber defence, as current Intrusion Detection Systems that rely solely on static attack signatures fail to protect against zero day attacks or advanced persistent threats (APTs). This thesis proposes a method of anomaly detection in networks that are part of large Systems of Systems SoS without prior definition of specific signatures of attacks, using the ML ML techniques. This study aims at improving the performance and precision of intrusion detection using benchmark classifiers including Random Forest, XGBoost, and Support Vector Machines (SVM), as well as benchmark datasets including NSL-KDD and UNSW-NB15. Pearson correlation factor is used in feature selection together with methods such as recursive feature elimination in order to refine inputs for enhancing the general model. The models are testing thoroughly for the accuracy, precision, recall, and F1-score that gives helpful information about discovering both known and new threats in cyber. Regarding critical issues, for example, high false-positive rates, or the need for further development of non-specific IDS models that would be able to address new threats so prevalent every time more sophisticated network structures are used, this research proposes the solutions for further development of highly effective, virtually non-resource-consuming IDS systems. The results have demonstrated that ML can be implemented as a strategic innovation in cybersecurity research and highlights a model for developing intelligent systems to counter contemporary threats
Reinforcing Security in Multi-Cloud Environment through Integrated Secure Data Transfer and Process Mining
No full textSecuring the multi-cloud environment has become essential in this world, where cybersecurity is a major concern. This research explores how data security could be improved in multi-cloud environments by integrating secure file transmission, dependable logging, and advanced analytical techniques like process mining. When data is being transferred within cross-cloud platforms, frequent cyberattacks and data breaches in cloud settings jeopardize data confidentiality and integrity throughout transfer operations, highlighting the need for stronger security measures. This study involves log analysis mechanisms that could identify deviations during file transfer. It will use a combination of process mining, visualization techniques, and secure data transfer protocols. This research focuses on automating and securing file transfers while ensuring compliance with security requirements by implementing a strong framework in a multi-cloud setting involving important platforms like AWS and Azure. In addition to addressing today’s security issues, this integrated strategy lays the groundwork for security solutions that can react to unexpected spikes in multi-cloud systems
Real-Time Detection of Social Engineering Threats in Social Media Posts
No full textSocial engineering attacks utilize online information, such as data from social media platforms, to indirectly obtain personal information. This method poses significant security risks by exploiting publicly available data to piece together sensitive information, making it harder for individuals to recognize and prevent such attacks. This research shows how we can predict susceptibility of social engineering by analysing social media posts. This research utilizes the YOLO (You Only Look Once) model, a real-time object detection system that processes an entire image in a single pass by predicting bounding boxes and class probabilities simultaneously, to detect objects such as laptops and dogs. Additionally, natural language processing is employed to analyse text for information like dates of birth and names, which are utilized in social engineering attacks. The application combines this analysis to provide users with immediate warnings by alerting them to potential security threats before they post on social media. This integrated approach, using both visual and text data, enhances the ability to predict social engineering attacks. This research contributes to cybersecurity by offering a proactive tool for protecting personal and corporate information shared on social media
Enhancing Network Security by Detecting Rogue Access Points using Ensemble Machine Learning Algorithms
No full textUnauthorised access points (APs) in wireless networks pose substantial threat to security, potentially resulting in data breaches and unauthorised access. Conventional security measures frequently fail to successfully identify these breaches, requiring the use of more sophisticated methods. considering the increasing reliance on wireless networks, especially in contexts where security is crucial, it is essential to create strong detection systems to identify unauthorised access points and protect the integrity of the network. This study utilised ensemble machine learning models, specifically Random Forest, Gradient Boosting, and AdaBoost, along with ANOVA feature selection, to identify rogue Access Points (APs) using the AWID dataset. The application of ensemble approaches, enhanced through the utilisation of Grid Search and cross-validation, greatly enhanced the accuracy of detection. The results indicated that the ensemble models surpassed the traditional models, with Gradient Boosting obtaining the highest level of accuracy. The SMOTE method was utilised to tackle the issue of data imbalance, resulting in improved model performance. However, the evaluation of certain metrics encountered difficulties due to constraints in computational resources. The results of this study add to the existing body of knowledge on network security by showing that ensemble learning approaches are more effective than traditional techniques such as KNN and SVM in identifying rogue access points (APs). The created model provides a reliable tool for network administrators, potentially reducing the likelihood of data breaches. Future research should prioritise the integration of threat intelligence to improve detection capabilities and investigate the system's capacity to recognise certain types of attacks
Securing Financial Transactions with Hybrid ECC and AES Encryption Algorithm
No full textCloud computing has achieved good attention among organization and user. By using it users can easily store their data directly on third party vendors and can enjoy the services, application and storage. This report is going to present the development and analysis of the SecureFinance system, which is a complex solution, and it has been designed to manage financial data in a secure way. The study will focus on the implementation of a hybrid cryptographic model that merges Elliptic Curve Cryptography (ECC) with Advanced Encryption Standard (AES). The primary aim is to increase data protection and performance in managing sensitive financial information as compared to traditional encryption methods. The report is going to find and evaluate the performance of ECC+AES against ECC+DES which shows its good performance and novelty as well. ECC provides strong security with smaller key sizes while AES gives strong encryption capabilities by making the hybrid approach highly good. The performance evaluation showed that ECC+AES delivers faster encryption and decryption times across various file sizes which also shows its practical advantages. The study includes deploying the SecureFinance system on Microsoft Azure in order to ensuring scalability and reliability in a cloud environment. It examines how well the system supports large-scale data management user access. By solving these key research questions related to security, performance and usability the report is going to contribute good data into the implementation of hybrid cryptographic techniques in financial systems
An intelligent Docker container-based solution with multiple IDS to filter DoS attack
No full textThis research focuses on enhancing the security of Docker container environments against denial-of-service (DoS) attacks through the deployment of multiple open-source Intrusion Detection Systems (IDS) tools. Docker containers are vulnerable to various DoS attacks that can severely impact system performance. This study integrates Snort, Suricata, and Zeek IDS tools within a Dockerized setup, using the ELK Stack for centralized log management and real time monitoring. The methodology involves simulating different types of DoS attacks, such as ICMP, TCP SYN, and UDP flood attacks to evaluate the detection capabilities of each IDS tool. The results demonstrate that a multi-layered defense strategy, combining the strengths of each tool significantly improves detection accuracy, scalability, and system efficiency. Snort was best in real time detection, Suricata managed high traffic volumes efficiently and Zeek provided in depth network analysis, making them a solution for securing Docker environments. The solution enhanced scalability and efficiency for DoS detection by using these tools together. Future work includes the integration of machine learning techniques to further enhance detection capabilities
Password storage protection using audio steganography with AES encryption
No full textSecuring end to end audio communication is very important but it is also having so many issues. This project presents a novel approach to secure password storage by combining audio steganography with AES encryption and SCRYPT hashing. The process starts with the user uploading an audio file and entering a password to be hidden. The original password is inserted into the audio file using Least Significant Bit (LSB) encoding by securing that the hidden data remains invisible. Following this the password experiences hashing using SCRYPT by generating a hash that serves as the key for AES encryption. The AES algorithm then encrypts the hashed password by producing ciphertext. The encryption and decryption processes have been measured for performance with results showing variations in encryption process time. The avalanche effect has been evaluated to test the sensitivity of the encryption algorithm to changes in the input password, revealing consistent high sensitivity with scores ranging from 48.23 to 51.62 which shows strong data encryption. For decryption the process includes retrieving the ciphertext from the steganographic audio which is going to extract and validate the hashed password and comparing it to the original hash. The decryption time is recorded and compared to the encryption time to evaluate performance. This approach is going to increase password security by combining steganography with advanced cryptographic techniques by having both secure storage and good data protection