NPS Australia Submission System
  • 0

    No products in the cart.

Paper Type: Application Paper

Road Sign Detection Using YOLO’s Latest Releases: An Evaluation Study of v8, v10, and v11

This paper presents a
comparative analysis of three recent YOLO variants—YOLOv8,
YOLOv10, and YOLOv11—evaluated on a traffic sign detection
task under variable real-world visual conditions. The nano
variant of each model was evaluated in terms of precision, recall,
mean average precision (mAP), training efficiency, F1-confidence,
and runtime speed. This study offers practical insights for
deploying object detection models in intelligent transportation systems, aiming to balance real-time performance with detection accuracy. The results indicated that YOLOv8 achieved the highest mAP (0.92), followed by YOLOv11 (0.908) and YOLOv10 (0.873). In terms of runtime performance, YOLOv8 and YOLOv11 demonstrated comparable speeds on the test data, whereas YOLOv10 required more time to complete the inference process

Modelling Moisture Recycling in the Sudd Wetland Using WAM-2layers for Sustainable Water Resource Management and Climate Resilience

This research provides the first detailed quantification of atmospheric moisture recycling in the Sudd Wetland, a globally significant yet understudied hydrological system. By integrating ERA5 reanalysis with the WAM-2layers model, it reveals the wetland’s substantial role in sustaining regional precipitation and its vulnerability to climate and land-use changes. The methodology offers a transferable framework for assessing land-atmosphere feedbacks in data-scarce regions, directly supporting sustainable water management, climate resilience planning, and transboundary governance in the Nile Basin and beyond.

Ecofiji Explorer: An App to Promote Eco-Tourism and Local Culture Through Digitalization

The paper contributes both theoretically (by linking ICT with sustainable tourism in a Pacific Island context) and practically (by developing and testing a real app prototype that empowers local communities and promotes eco-tourism).

Bula Patrol: A Comprehensive Monitoring System for Addressing Taxi Driver Vulnerabilities in Fiji

The research clearly presents the critical issues faced by taxi drivers in Fiji, outlines the proposed solution (the Smart Taxi Monitoring System), and highlights the potential benefits of implementing this system. It succinctly conveys the urgency of the problem and the innovative approach taken to address it, making it relevant to stakeholders in transportation safety, technology, and public policy.
Additionally, the emphasis on real-time monitoring and predictive analytics demonstrates a forward-thinking solution that aligns with current trends in technology and safety management. This combination of practical implications and technological advancement makes the content suitable for a broader audience interested in industry improvements and public safety.

Forecasting of maximum temperature using ETS, ARIMA and Random Forest models: A case study for Karachi, Pakistan
An AI-Enabled Centralized Monitoring System to Predict SME Inventory Level

I. To utilize the historical data, and predict market needs in a dynamic environment to maintain inventory level. (To develop a data-driven inventory management system)
II. To observe, track, and learn about product movement by implementing an AI-powered system. (To implement an AI-powered product movement tracking system).
III. To optimize the accuracy of prediction for market demand forecasting. (To improve and refine the prediction model).

Deployment of AI Models and a TDA Mapper Algorithm to Enhance Clinical Decision-Making

This research presents a novel Clinical Decision Support System (CDSS) that integrates advanced AI technologies, including a hybrid model for COVID-19 diagnosis and a custom GPT Assistant, into a scalable, real-world tool for resource-limited healthcare settings. The CDSS enhances the quality of care by providing accurate, timely diagnostics and decision support, while also improving response efficiency through real-time monitoring and predictive analytics. The system’s adaptability, supported by open-source platforms like R Shiny, and its potential for wide adoption, particularly after a planned pilot and impact evaluation, highlight its significance in advancing healthcare delivery where resources are scarce. The deployment of this AI-driven CDSS has the potential to transform clinical decision-making, improve healthcare delivery, and enhance the overall response to public health emergencies. The structured pilot and impact evaluation will provide critical insights into the system’s effectiveness in real-world settings, paving the way for broader adoption and sustained improvements in healthcare quality and efficiency.

Topological Machine Learning: Integrating Topological Data Analysis with Machine Learning to Enhance Breast Cancer Classification

Novel TML Approach: Introduces Topological Machine Learning (TML), integrating Topological Data Analysis (TDA) with Machine Learning (ML) to enhance breast cancer classification.

Improved Accuracy: Demonstrates that TML significantly improves classification accuracy on the Wisconsin Breast Cancer (WBCD) dataset compared to traditional methods and standalone ML techniques.

Feature Evaluation: Provides a detailed assessment of topological features such as cluster means, node features, and link features, highlighting their impact on model performance.

Practical Insights: Offers valuable insights into the integration of topological features for better diagnostic accuracy, with implications for improving breast cancer classification and patient outcomes.

Is Benford’s Law Based Detectors Effective for GAI Generated Images?

Benford’s Law is used in image forensics. We tested GAI generated images to see whether BL is able to detect artificially generated images. The experiments show that only about 60\% of the images can be detected using a simple similarity threshold.

ActJOLO: Action Recognition Guided by Actionlets Using Joint Lightweight Optical Flow Information

In this study, we propose a novel method, named ActJOLO, which builds upon the existing JOLO model by incorporating an advanced self-supervised learning technique as an upstream guide for posture recognition. Our approach emphasizes the analysis of high-intensity motion features within the human body, thereby enhancing the efficiency of action modeling.
Experimental results on the NTU RGB+D dataset demonstrate that our framework improves processing speed compared to the original model, while maintaining high ccuracy. This work offers a new perspective on skeleton-based human action recognition and highlights its potential for deployment on low-performance processors.