International Journal of Innovations in Science & Technology
Not a member yet
813 research outputs found
Sort by
Leveraging Generative AI to Learn Impact of Climate Change on Buildings Urban Areas
Get PDFClimate change, global warming, and pollution are intensifying daily. As urbanization increases, understanding the reciprocal impact between buildings and the environment becomes increasingly important. While most research on building monitoring through the Internet of Things (IoT) emphasizes energy consumption and data collection, it often overlooks the effects of outdoor environmental factors on buildings and vice versa. Additionally, existing studies frequently lack detailed reports that clarify their findings. This work aims to expand our understanding of environmental influences on buildings, indoor environments, and residents. It also seeks to generate comprehensive reports on these impacts, providing actionable recommendations to mitigate and minimize them through the use of Generative Artificial Intelligence. Specifically, we fine-tuned Large Language Models (LLMs) such as Generative Pre-trained Transformer 2 (GPT-2) and Large Language Model Meta AI 2 (LLAMA2-7b), using the Nous Research LLAMA2-7b-hf version from Hugging Face, on a custom dataset compiled from diverse online sources. Our research examines the effects of environmental factors, including temperature, humidity, and air quality, on urban buildings and indoor environments, with these models generating reports that offer practical recommendations. The generated reports offer a clear understanding of environmental impacts on buildings and suggest strategies to minimize these effects. These insights are intended to support effective urban planning and sustainable development. By implementing these recommendations or best practices, we can enhance indoor environmental quality while reducing contributions to global warming. Future work will involve continuous monitoring of buildings\u27 indoor environments, energy consumption, and greenhouse gas (GHG) emissions, further reducing GHG emissions and addressing global warming
Predictive Analysis and Email Categorization Using Large Language Models
Get PDFWith the global rise in internet users, email communication has become an integral part of daily life. Categorizing emails based on their intent can significantly save time and boost productivity. While previous research has explored machine learning models, including neural networks, for intent classification, Large Language Models (LLMs) have yet to be applied to intent-based email categorization. In this study, a subset of 11,000 emails from the publicly available Enron dataset was used to train various LLMs, including Bidirectional Encoder Representations from Transformers (BERT), Distil BERT, XLNet, and Generative Pre-training Transformer (GPT-2) for intent classification. Among these models, Distil BERT achieved the highest accuracy at 82%, followed closely by BERT with 81%. This research demonstrates the potential of LLMs to accurately identify the intent of emails, providing a valuable tool for email classification and management
Exploring the Efficacy of CNN Architectures for Esophageal Cancer Classification Using Cell Vizio Images
Get PDFEsophageal cancer, as with the global burden of disease, is usually due to Barrret\u27s esophagus and gastroesophageal reflux disease. Fortunately, the disease is amenable to early detection; however, early diagnosis has been complicated by the limitations of the existing diagnostic technologies. To address this problem, a new Convolutional Neural Network and ResNet50 architecture are presented in this study to aid esophageal cancer diagnosis through the classification of Cell Vizio images. This diagnosis is made by the deep learning architecture which does tissue classification into four categories thus improving the diagnostic sensitivity. For model training and testing preoperative perforations in sixty-one patients, 11,161 images were used. Data augmentation and normalization techniques were also performed on the images to help improve the outcome. Our training accuracy reached an impressive 99% 12, while our final f1 score was 93.05%. Our Res Net 50 model obtained an F1 score of 93.26%, precision of 94.05%, recall of 93.52 %, and validation accuracy of 93.32 %. These results indicate how well our deep learning-based technique can be used as a quick, non-embolic, accurate method for early detection of esophageal carcinoma
Fabrication of Smart Syringe Infusion Device: A Solution for Healthcare Industry
Get PDFAccurate medication delivery is essential for patient outcomes in intensive care units, where precision in drug delivery is crucial. In order to address the need for increased accuracy and efficiency in workflow, this study proposes a semi-automated smart syringe infusion device with a unique refill mode integrated with electronic health records (EHR). The device was tested in both manual and virtual control modes, with a stepper motor-driven syringe used for precise fluid infusion. The refill mode was evaluated based on its ability to automate the refilling procedure. The results showed precise dosage control in a variety of scenarios, with minimal discrepancies between the desired and actual amounts. The refill mode effectively automated the withdrawal and refilling processes, lowering human error and increasing efficiency. Additionally, the device\u27s effortless interface with EHR systems streamlined the documentation process, enabling real-time data logging and enhancing workflow. This device offers a dependable, cost-effective solution for improving medication delivery, making it a valuable tool in healthcare, particularly in resource-limited environments
Event-Based Vision for Robust SLAM: An Evaluation Using Hyper E2VID Event Reconstruction Algorithm
Get PDFThis paper investigates the limitations of traditional visual sensors in challenging environments by integrating event-based cameras with visual SLAM (Simultaneous Localization and Mapping). The work presents a novel comparison between a visual-only SLAM implementation using the state-of-the-art HyperE2VID reconstruction method and conventional frame-based SLAM. Traditional cameras struggle in low dynamic range and motion blur scenarios, limitations that are addressed by event-based cameras, which offer high temporal resolution and robustness in such conditions. The study employs the HyperE2VID algorithm to reconstruct event frames from event data, which are then processed through the SLAM pipeline and compared with conventional frames. Performance metrics, including Absolute Pose Error (APE) and feature tracking performance, were evaluated by contrasting visual SLAM implementations on reconstructed images against those from traditional cameras across three event camera dataset sequences: Dynamic-6DoF, Poster-6DoF, and Slider depth sequence. Experimental results demonstrate that event-based cameras yield higher-quality reconstructions, significantly outperforming conventional cameras, especially in scenarios marked by motion blur and low dynamic range. Among the tested sequences, the Poster-6DoF sequence exhibited the best performance due to its information-rich scenes, while the Slider depth sequence faced challenges related to drag and scaling, as it lacked rotational motion. Although the APE values for the Slider depth sequence were the lowest, it did experience trajectory drift. In contrast, the Poster-6DoF sequence displayed superior overall performance, with reconstructions closely aligning with those produced by conventional camera-based SLAM. The Dynamic-6DoF sequence showed the poorest performance, marked by high absolute pose error and trajectory drift. Overall, these findings highlight the substantial improvements that event-based cameras can bring to SLAM systems operating in challenging environments characterized by motion blur and low dynamic ranges
Semantic Segmentation Based Lightweight Lane Detection Network (LW Net) for Intelligent Vehicles
Get PDFA novel lane detection system is proposed for intelligent vehicles. A key feature of this system is its lightweight design, which requires less computational power. Our lightweight network (LW Net) for semantic segmentation comprises convolutional and separable convolutional layers. We designed a total of six lightweight encoder models (LW Net-A, LW Net-B, LW Net-C, LW Net-D, LW Net-E, and LW Net-F), each paired with matching decoders. The first group of three models is based on depth D1, while the remaining models are based on depth D2. In these models, convolutional layers are either fully or partially replaced by separable convolutional layers. The lightweight network LW Net-A achieved an 88% reduction in training parameters, along with a 2.45% increase in test accuracy compared to the benchmark Seg Net model. Meanwhile, LW Net-F attained a 2% increase in test accuracy and a remarkable 94% reduction in training parameters compared to the benchmark Seg Net model. Overall, the proposed models are less computationally demanding than other benchmark networks, without compromising the pixel accuracy of the semantic model
Development of Narrow Band Internet of Things Testbed for Proximity Services
Get PDFIn this paper, we present the results of this deployment, indicating that the NB-IoT device successfully connects and communicates with the eNodeB. Session logs show that the EPC effectively initiates the session and authenticates the user equipment (UE). Additionally, the eNodeB establishes a successful connection with the UE based on the parameters defined by the EPC. Wireshark traces demonstrate that the UE can transmit data to the server via an internet connection, with an average latency of 40 ms. Through this work, we explore the benefits of proximity services for NB-IoT networks, providing a valuable platform for experimental testing and prototyping. With the evolution of the Internet of Things (IoT), Narrowband IoT (NB-IoT) has emerged to provide IoT connectivity over existing cellular networks, utilizing limited resources while facing an increased risk of outages at the cell edge. In this work, we developed a testbed for NB-IoT systems to implement the innovative idea of enabling proximity services. This allows devices beyond the coverage area of the eNodeB to transmit their data using a network relay. The testbed is constructed with a software-defined radio functioning as the eNodeB, which wirelessly connects to the NB-IoT node at the front end and to the Evolved Packet Core (EPC) at the back end. The EPC is implemented on a Linux machine using open-source software, while the NB-IoT node is realized with commercially available devices
A Novel Guard Zone Based Multiple Access Protocol for Autonomous D2D Cellular Communication
Get PDFIn today\u27s interconnected and digitally driven era, Device-to-Device (D2D) communication has emerged as a transformative paradigm, enabling direct and efficient interactions between nearby devices with or without traditional network infrastructure. This research introduces a novel communication protocol that autonomously facilitates D2D communication in network-constrained environments, maximizing node engagement and ensuring reliable data transmission while addressing the inherent challenges faced by existing Medium Access Control (MAC) protocols, such as Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA). The study specifically focuses on developing a protocol that enables real-time information sharing among neighboring nodes without network assistance. To evaluate the protocol\u27s performance, comprehensive simulations were conducted using MATLAB, assessing its effectiveness in increasing the number of active nodes and reducing collision probability. The results demonstrate that the proposed protocol significantly decreases interference while enhancing data throughput and energy efficiency, achieving a 15% increase in active node pairs and a notable 4% reduction in collision probability. By minimizing contention overhead and optimizing data transmission, the protocol effectively lowers latency, ensuring reliable communication in environments lacking network support. Furthermore, it conserves energy by reducing idle listening, thereby extending battery life and promoting sustainable wireless communication systems. This research provides a robust solution to enhance D2D communication in isolated environments, paving the way for more resilient wireless ecosystems
Leveraging Cryptographic Primitives of Blockchain for Trust in Smart Systems: -
Get PDFCalculating and maintaining trust using Hyperledger Fabric in smart systems plays a vital role in mitigating various trust-related attacks. Current smart systems encounter several challenges, including dependence on centralized trust authorities, which are prone to attacks and present single points of failure, as well as the need to maintain user privacy while establishing trust. Ensuring data integrity and authenticity is equally crucial. In these systems, nodes assess the trustworthiness of other nodes based on their experiences and recommendations. However, trust calculations can be vulnerable to integrity attacks from malicious nodes, such as bad-mouthing and ballot stuffing. To address these threats, trust can be calculated and securely stored on the blockchain. We selected Hyperledger Fabric as the blockchain framework and conducted a prototype implementation of trust calculation on a reduced scale involving 10 nodes. Hyperledger Fabric, being a private, permissioned blockchain, is suitable for decentralized trust calculations and storage in smart devices. We simulated a healthcare scenario within an HLF network, demonstrating secure trust calculation among IoT devices. The results indicate that leveraging the cryptographic properties of blockchain significantly enhances the overall security and trustworthiness of smart systems
Integrating LLM for Cotton Soil Analysis in Smart Agriculture System
Get PDFCotton is a critical crop for the agricultural economy, with its productivity closely tied to soil quality, particularly soil nutrient levels and pH. Monitoring and optimizing these properties is essential for sustainable cotton cultivation. This study proposes using fine-tuned Large Language Models (LLMs)—specifically GPT-2 and LLaMA-2—to automate soil analysis and produce detailed soil reports with actionable recommendations, addressing the limitations of traditional machine learning models in this context. A custom dataset was created by extracting key information from cotton-specific resources, focusing on soil nutrient interpretation and recommendations across different growth stages. Fine-tuning was applied to GPT-2 and LLaMA-2 models (specifically, the Nous Research version LLaMA2-7b-hf from Hugging Face), enabling them to generate data-driven reports on cotton soil health. The fine-tuned GPT-2 model achieved a training loss of 0.093 and an evaluation loss of 0.086, outperforming LLaMA-2, which had a training loss of 0.033 and an evaluation loss of 0.25. Evaluation with BERT Score showed that GPT-2 scored average Precision, Recall, and F1 scores of 0.9284, 0.9308, and 0.9296, respectively, highlighting its superior report accuracy and contextual relevance compared to LLaMA-2. The generated reports included soil properties and actionable nutrient management recommendations, effectively supporting optimized cotton growth. Implementing fine-tuned LLMs for soil report generation enhances nutrient management practices, contributing to higher yields and more sustainable cotton farming