International Journal of Innovations in Science & Technology
Not a member yet
813 research outputs found
Sort by
On Evaluation of discrete RL agents for Traffic Scheduling and Trajectory Optimization of UAV-based IoT Network with multiple RIS unit
Get PDFUnmanned Aerial Vehicles (UAVs) have been very effective for data collection from widely spread Internet of Things Devices (IoTDs). However, in case of obstacles, the Line of Sight (LoS) link between the UAV and IoTDs will be blocked. To address this issue, the Reconfigurable Intelligent Surface (RIS) has been used, especially in urban areas, to extend communication beyond the obstacles, thus enabling efficient data transfer in situations where the LoS link does not exist. In this work, the goal is to jointly optimize the trajectory and minimize the energy consumption of UAVs on one hand and satisfy the data throughput requirement of each IoTD on the other hand. As it is a mixed integer non-convex problem, Reinforcement Learning (RL); a class of Machine Learning (ML), is used to solve it, which has proven to be computationally faster than the conventional techniques to solve such problems. In this paper, three discrete RL agents i.e. Double Deep Q Network (DDQN), Proximal Policy Optimization (PPO), and PPO with Recurrent Neural Network (PPOwRNN) are tested with multiple RISs to enhance the data transfer and trajectory optimization in an Internet of Things (IoT) network. The results show that DDQN with multiple RIS is more efficient in saving communication-related energy, while a single RIS system with the PPO agent provides more reduction in the UAV’s propulsion energy consumption when compared to other agents
Use of Artificial Intelligence in Ethereum Forecasting: The Deep Learning Models RNN and CNN with Ensemble Averaging Technique
Get PDFIn the fast-evolving cryptocurrency market, accurately predicting Ethereum prices is crucial for investors, traders, and financial analysts. Traditional machine learning (ML) models often struggle to capture the market\u27s complex dynamics due to their inability to consider all influencing factors. This study introduces an advanced ensemble machine learning approach to enhance Ethereum price prediction accuracy. By combining the strengths of Bi-directional Long Short-Term Memory (Bi-LSTM) and Convolutional Neural Network (CNN) models, our ensemble averaging method compensates for individual model weaknesses, improving forecast reliability and precision. Results show that our ensemble model offers significant advantages, particularly in terms of generalizability and resistance to overfitting with LSTM and CNN models and this technique is offering a more effective tool for navigating cryptocurrency market complexities. This research highlights the importance of ensemble learning in financial forecasting and provides a practical framework for developing superior predictive models. “Moreover, This study explores an advanced ensemble machine learning approach to enhance Ethereum price predictions, combining the strengths of Bi-directional Long Short-Term Memory (Bi-LSTM) and Convolutional Neural Network (CNN) models. While Bi-LSTM individually exhibits slightly higher performance in our tests, the ensemble method demonstrates enhanced stability and reliability, making it a valuable tool for navigating the unpredictable dynamics of the cryptocurrency market. We found that Bi-LSTM is good on its own, but the balanced approach of the ensemble model is far better, especially when it comes to generalizability and overfitting resistance. Insights into creating flexible and trustworthy prediction models are provided by this study, which highlights the possibilities of ensemble learning in financial forecasting
Game Brains: NPCs Intelligence Using Neural Network Brains
Get PDFThis paper aims to develop the foundational knowledge about the Unity game development engine embedded with AI for the development of a hyper-casual game that has intelligent NPCs, which operate strategically in the environment. The targeted audience comes in the class of those who are pursuing their career in the niche of AI game development and enhancing the gaming experience for single-player game users. Using Unity Engine and Python, Curriculum learning and self-learning experiments were conducted to test the AI game. Moreover, in this paper, different reinforcement learning methods have been discussed, which have been implemented in the game that produces the optimal results for the behavior of NPCs. Hence, this paper tends to represent a glimpse into the future perspective of the gaming industry in hyper-casual gaming platforms
Low-Cost Smart Metering Using Deep Learning
Get PDFUtility services like electricity, water, and gas are essential for modern living, and their demand has been rising worldwide. However, traditional manual meter reading is a standard procedure for billing purposes. This is not only labor and time-intensive but also prone to mistakes, which results in incorrect billing and revenue losses. In the era of advanced AI, leveraging cutting-edge technology to automate meter readings has become increasingly viable. However, Existing AI-based meter reading systems have limitations in detecting and recognizing meters from a distance. This research addresses these problems by presenting a novel system that utilizes the YOLOv8 model to detect meter screens from a distance. In addition, the system uses a fine-tuned Paddle OCR to recognize meter readings. A Novel dataset curated for the meter screen detection, recognition, and end-to-end OCR tasks related to electricity, gas, and water utility meters has been presented, containing up to 8,044 images. The proposed system was trained and extensively tested on the proposed dataset to gauge its performance. The system achieved an exceptional mean Average Precision (mAP) of 0.995 for both analog and digital meters on the detection task; furthermore, the system achieved an accuracy of 96.92% in the recognition task, which is 70% better than the accuracy of Pre-trained Paddle OCR. Moreover, an all-encompassing evaluation that combines detection and recognition using Paddle OCR and YOLOv8, i.e., the end-to-end OCR task, achieved an accuracy of 97.8%. Lastly, the system achieved an inference speed of up to 6 frames per second, guaranteeing real-time effectiveness
Using Spatial Covariance of Geometric and Shape Based Features for Recognition of Basic and Compound Emotions
Get PDFIntroduction. Compound emotion recognition has been an emerging area of research for the last decade due to its vast applications in surveillance systems, suspicious person detection, detection of mental disorders, pain detection, automated patient observation in hospitals, and driver monitoring.
Objectives: This study focuses on emotions, highlighting the fact that the existing knowledge lacks adequate research on compound emotions. This research work emphasizes compound emotions along with basic emotions.
Novelty Statement: The contribution of this paper is three-fold. The study proposes an approach relying on geometric and shape-based features using SVM and then fusing the obtained geometric and shape-based features for both basic as well as compound emotion recognition.
Materials and Method: This study provides a comparison with six state-of-the-art approaches in terms of percentage accuracy and time.
Dataset: The experiments are performed on a publicly available compound emotion recognition dataset that contains images with facial fiducial points and action units.
Result and Discussion: The results show that the proposed approach outperforms the existing approaches. The best accuracy achieved is 98.57% and 77.33% for basic and compound emotion recognition, respectively. The proposed approach is compared with existing state-of-the-art deep Neural Network architecture. The comparison of the proposed approach has been extended further to various existing classifiers both in terms of percentage accuracy and time.
Concluding Remarks: The extensive experiments reveal that the proposed approach using SVM outperforms the state-of-the-art deep Neural network architecture and existing classifiers including Naive Bayes, AdaBoost, Decision Table, NNge, and J48
Applications of RS & GIS for Tsunami and Sea Surges Risk Assessment Along the Coast of Karachi
Get PDFCoastal areas are vulnerable to various hazards, such as storm surges, inundation from sea-level rise or coastal flooding, tsunamis, and more. The situation becomes particularly disastrous if the coast is densely populated and highly developed. Pakistan has a coastline stretching 1,046 km, with Karachi being the most developed part. The Karachi coast faces frequent storms during the monsoon season and is also threatened by rising sea levels in the coming years. Additionally, Pakistan\u27s coastline is near the boundaries of two major tectonic plates—the Indo-Australian and Eurasian plates—as well as two minor plates, the Arabian and Iranian plates. In the event of a major earthquake in the Arabian Sea, a tsunami could pose a significant threat, potentially engulfing important and densely built-up commercial, residential, industrial, and sensitive military areas. This study aims to analyze potential losses due to inundation of the Karachi coast using Remote Sensing and GIS techniques
Impact Assessment of Agro-Meteorological Drought Using Geo-Spatial Techniques. A Case Study of Southeastern Sindh-Pakistan
Get PDFThe frequency of droughts is increasing as global temperatures rise. To effectively monitor drought conditions, it is crucial to use the appropriate index. In this study, the Standardized Precipitation Index (SPI) and Reconnaissance Drought Index (RDI) were applied to evaluate droughts. The tool "DrinC" was used to calculate the RDI for 3-, 6-, and 12-month periods (Oct-Dec, Oct-March, and Oct-Sept) from 1981 to 2020. RDI values between -1.0 and -2.5 indicated moderate to extreme droughts across all districts. The RDI for 3, 6, and 12 months highlighted significant drought years, including 1984, 1992, 1994, 2010, 2011, 2015, and 2019, showing reduced productivity during these periods. Dry conditions were prevalent at most stations between 1981 and 2020. In South-Eastern Sindh, Pakistan, this study also assessed changes in Land Surface Temperature (LST), Normalized Difference Vegetation Index (NDVI), and Soil Moisture Index (SMI) over the last four decades (1981-2020). Satellite data analysis showed that NDVI peaked in 1988 (+0.53) and hit its lowest in 2021 (+0.48). Similarly, SMI ranged from +1.1 in 1988 to +0.98 in 2021, while LST increased from 35.1°C in 1988 to 53.4°C in 2021. A negative correlation between SPI and RDI was observed through linear regression, confirming the effectiveness of both indices in assessing drought severity. These findings can inform the development of drought preparedness plans, helping to mitigate the impact of drought on various economic sectors
Multi-Digit Number Recognition System: Single-Digit CNNs for Multi-Digit Detection and Recognition Using MNIST Dataset
Get PDFThis research focuses on creating a deep-learning model for identifying multi-digit numbers, which addresses the critical demand for accuracy in real-world applications. The study presents novel approaches to multi-digit recognition, providing a thorough resolution to an unsolved problem in the field of computer vision. In order to improve model generalization, the study makes use of convolutional neural networks (CNNs) that were trained on the MNIST dataset and augmented with rotation and scaling approaches. Multi-digit number prediction is a multi-step process. detection to isolate each digit. Each digit is then clipped and stored separately with its own label. Subsequently, the algorithm predicts the digit for each cropped picture and saves them. This method is repeated for all identified contours, with each predicted digit concatenated to get the final multi-digit prediction. Finally, the projected multi-digit sequence is compared to the ground truth for assessment. The CNN achieves remarkable training and validation accuracies of 99.60% and 99.28%, proving its ability to recognize multi-digit numbers. This study emphasizes the importance of advanced methods in developing deep learning models for multi-digit recognition, which promise enhanced automation and efficiency across a variety of digital technology industries
Cause and Damages Assessment of 2022-Flood in Khyber Pakhtunkhwa, Pakistan
Get PDFFloods are among the most devastating hazards, occurring globally and impacting many regions annually. Pakistan is frequently affected by floods, including the significant floods of 2022 in Khyber Pakhtunkhwa (KP). This study assesses the causes and damages of the 2022 floods in KP using data from NASA Worldview and USGS, complemented by Geographical Information System (GIS) analysis. The study considers the role of climate change and the topography of KP in making it prone to floods. It examines weather patterns, environmental factors, and local vulnerabilities that contributed to the floods, as well as the extent of damage to communities, infrastructure, and the environment. Flood and precipitation data were collected from two satellites and analyzed using ArcGIS. The study identified massive rainfall and increased temperatures as the primary causes of the flood. Significant damage was recorded in District Dera Ismail Khan, followed by Tank and Swat. The floods resulted in approximately 300 fatalities across various districts of KP and caused total economic losses estimated at 201,414 million Pakistani rupees. Public sector losses were estimated at 121,283 million PKR, with house damages amounting to 23,780 million PKR. The peak flooding occurred in August during high rainfall. Understanding the root causes and damages of the 2022 KP flood is crucial for developing effective prevention and mitigation plans, as well as for assessing the impact on communities, infrastructure, and the environment. This study provides critical insights and comprehensive data to inform disaster management and policy-making for future resilience. Its novelty lies in its exclusive focus on the 2022 KP floods, a topic not previously studied in detail. In conclusion, the research effectively analyzes the causes and assesses the damages of the 2022 Khyber Pakhtunkhwa flood, offering essential insights for improving flood management strategies
New Bovid (Artiodactyla) Fossils from the Siwaliks of Pakistan: Reviving a Lost World
Get PDFThis research investigates new fossil specimens of bovids (Artiodactyla) from the Dhok-Pathan Formation in the Siwalik region of Pakistan, a crucial site for understanding South Asia\u27s paleoecology and evolutionary history. This study provides new insights into the taxonomy and diversity of Siwalik bovids, addressing gaps in the fossil record and contributing to a more comprehensive understanding of their evolutionary relationships. In this paper, new dental elements of bovids were recovered (seven specimens were collected) comprised of both upper and lower dentitions have been recovered from the most fossiliferous sites i.e., Dhok Pathan and Hasnot villages, in Potwar foreland basin of Himalayas in Pakistan. On the basis of comparative morphology and the precise measurements of these specimens refer to the mandible of Selenoportax vexillarius, rest of all are molars and premolars of Pachyportax latidens, Pachyportax nagrii and Kobus porrecticornis. All the new dental material is documented in this research belongs to Dhok Pathan Formation of upper Siwaliks Group. The stratigraphic layers, encompassing various depositional environments such as river channels and floodplains, provide insights into the chronological and environmental contexts of the fossil assemblages. Comparative analysis with fossils from other regions ensures the accuracy and relevance of the findings, contributing to a refined understanding of the evolutionary history and biogeographic patterns of these species. The study also documents the paleoecology of the region, indicating a grassland and woodland biome that supported diverse bovid species. These findings underscore the Siwalik region\u27s significance as a key site for studying the evolutionary history of Artiodactyla and provide valuable data for future paleontological and conservation studies. The examined fauna suggests a vast and an open landscape with intermittent dry and flood seasons, creating a mosaic of ecotonal habitats with numerous niches. This research enhances our knowledge of the Siwalik region\u27s past biodiversity and environmental changes, emphasizing the need for continued exploration and analysis of its fossil record