UTAR Institutional Repository (Universiti Tunku Abdul Rahman)
Not a member yet
6132 research outputs found
Sort by
Development of a cross-platform mobile application for monitoring palm oil mill (POM) processes
No full textThis collaborative project with Novaflow Engineering Sdn. Bhd. aims to overcome the limitations of the current mobile application used for monitoring palm oil mill (POM) processes. The existing system faces challenges in terms of viewing the real-time data and receiving alarm notifications, which affect operational efficiency and safety. The proposed solution for this project involves designing and developing a cross-platform mobile application that shall provide a visual representation of the POM process layout. The application shall display real-time data updates, alarm notifications for critical conditions, and improvements in the user interface for enhanced usability. Additionally, filtering graph features is added to facilitate easy data analysis and comparison. The system also integrates with Influx DB, the database currently used by the company to store data. The proposed solution ensures more efficient monitoring, enhanced safety, and improved operational performance. In conclusion, the project delivers a practical and efficient solution that meets the operational needs of Novaflow Engineering and provides a robust, cross-platform application to support ongoing improvements in the company’s POM operations.
Keywords: palm oil mill monitoring; cross-platform mobile application; real-time data visualization; alarm notification system; industrial process automation
Subject Area: TS155–194 Production management. Operations managemen
Linen monitoring using rfid for UTAR Hospital
Get PDFThis project explores the application of Radio Frequency Identification (RFID) technology to enhance linen management in a hospital. It introduces a semi-automated system to enhance the efficiency and accuracy of linen handling and tracking. The system integrates RFID readers and a backend server to track and identify linen items in real time. It has a simple interface that makes the linen handling process easy to use, even for users with different educational levels.
Throughout the development process, the project progressed from literature review and research to prototype implementation and testing. Despite the difficulties in the form of hardware limitations and the limited resources available, an RFID-based linen monitoring fully functional prototype was established. The system is able to read RFID tag data, process it through the server, and depict key information to enable improved monitoring and decision-making.
Testing confirmed that the system accurately detects, tracks, and manages linen items in a controlled environment, thereby confirming its feasibility for real-world deployment. This achievement demonstrates a practical and scalable solution for improving hospital linen operations. The project is expected to be completed within eight months to support timely deployment in a healthcare setting
Developing a fast scam prevention mobile application: large language models
Get PDFScams are on the rise and constantly evolving. Threat actors have abused the rise of LLM to ease the process of creating deception information for scams. Manually flagging scam information is tedious and needs to be faster to counter the rapid growth of scam cases. Therefore, this project proposes developing a "Sentinel," a real-time scam detection system leveraging a large language model (LLM) for enhanced analysis of scam audio and text messages. The strategy is to build a mobile application that automatically captures the user's text and audio input and then utilizes Google's Gemini LLM for content analysis which maximizes the flexibility of the scam detector deal with new contents smoothly. After the complete LLM analysis is ready, the application will alert the user regarding the content analysis. The project considers the importance of the user's privacy by building an active application where user content analysis will only be done upon request, ensuring that the user has full control over when and how their data is analysed. The developed application would be capable of being implemented in the majority of Android devices with minimized performance hit on lower-end smartphones
Development of an interactive automatic smart bin system
Get PDFThe increasing population in Malaysia causes unsustainable waste generation. Malaysia must improve traditional waste management and reduce waste disposal in landfills. Nearly half of Malaysians are aware of environmental issues and recycling, as well as knowing waste segregation. Concurrently, traditional waste collection is low productivity compared to checking and emptying bins individually. The interactive automatic smart bin system is suggested as the solution. The smart bin can increase user engagement and encourage proper waste segregation behaviours, moreover, stays available by remotely monitoring the bin status and sending to the waste collectors. The proposed system classifies the class of waste based on images captured. The bin senses the type of recyclable waste such as paper, plastic, and metal, as well as the trash. Then, automatically sort the waste to dispose of in the appropriate bin. Thus, users don't have to worry about which bins to throw. The system is integrated with Wi-Fi to communicate with the bins. When the bin is full, the system will send a notification to waste collectors. Therefore, the waste collection uses less manpower, fuel cost, and time consumption
Audio files comparator using wavelet transform and similarity metrics
Get PDFThis project is a development-based project revolving around signal processing. The aim of this project is to develop a program that utilizes continuous wavelet transform (CWT) for audio similarity recognition. Its primary objective is to identify the similarities among audio files with different information such as file names or formats.
In today’s diverse musical landscape, songs undergo various interpretations, covered in different languages, or rendered using a myriad of instruments. Compositions may span the spectrum, ranging from performances with real musical instruments to those composed solely of synthesized sounds, typically electronic dance music (EDM).
Furthermore, songs exhibit versatility in their presentation, ranging from vocal renditions accompanied by instruments to whistling, humming or acapella performances. The evolution of music has also fostered the emergence of mashups and remixes, where distinct tracks seamlessly blend together to create new compositions. Despite these variations, the tunes or pitches of songs remain recognizable to the human ear and even audio detection algorithms. With the proliferation of digital music, people download songs from music applications or the internet, whether for personal listening in vehicles or to play in parties. However, these downloaded songs may vary depending on their file names and formats. Consequently, this project aims to identify identical or akin songs with various information and display out the percentage of differences between the audio files. The project’s methodology centres on Python programming, where comparisons of audio similarities will be conducted
Fish pellet measurement system for food industry
Get PDFThis project aims to address the need for precise measurement of fish pellets in the industrial sector by leveraging technology. This project will involve several fields such as computer vision, embedded systems and deep learning. Before that, manpower is required to measure the fish pellets individually. By using this traditional method, the accuracy and efficiency are low. Therefore, a new method using technology will be introduced to solve this problem. The core technology used in this project is computer vision and deep learning. Initially, a camera will be set up to capture the fish pellet. Then, the image will be processed in a trained model to detect the fish pellet. Then, an algorithm will be used to determine the fish pellet's diameter based on the result of the detection. To improve the consistency, Raspberry Pi will be chosen as the CPU of this project. The camera will be interfaced to it, and the trained model will be imported into it. A user-friendly GUI will also be provided to display the output information of the system. Python will be selected as the programming language in this project due to its extensive library support, such as OpenCV for computer vision and TensorFlow for deep learning. As a result, the GUI should perform the fish pellet detection and diameter calculation, which has a bounded box and label of the diameter value on each fish pellet as the output. In conclusion, this project will provide a more efficient solution than the traditional method
Exploring the potential of using aruco markers to monitor fish feeding status
Get PDFEfficient feeding management is a cornerstone of sustainable aquaculture, directly influencing fish growth, health, and resource utilization. Traditional feeding methods, which rely on manual observation to determine satiety, are labour-intensive, subjective, and prone to human error—often resulting in overfeeding and operational inefficiencies. This project presents a novel approach for monitoring fish feeding status by leveraging ArUco marker tracking. Pose estimations of floating markers are analysed to extract movement intensity, which is then interpreted using a time-series LSTM classification model to detect fish activity and infer satiety levels. The system was developed using a combination of Python, Keras, and OpenCV, and deployed in a real aquaculture setting using red hybrid tilapia (Oreochromis sp.). A web-based interface provides real-time pose data, fish activity classification, feeding recommendations, and status tracking. Model performance was validated through cross-validation and real-world testing, achieving high accuracy and practical reliability. Beyond monitoring fish feeding status, the system also detects air pump operation and tracks water level variations, offering a broader view of tank conditions. It supports multi-tank monitoring using a single camera, making the solution cost-effective, scalable, and non-invasive. The results affirm the system’s potential to improve feed management, reduce labour dependency, and support more intelligent and sustainable aquaculture practices
Real-time money counting app for visually impaired
Get PDFVisually impaired individuals usually need assistance from others to perform daily activities such as grocery shopping, reading documents, and recognizing banknotes due to their limited vision. Low vision or blindness causes their daily life activities to become time-consuming, especially dealing with financial transactions because they are not able to determine the amount of money that they are handling especially if the banknote is old, worn, or even torn. Even though efforts are made to print Braille on the banknotes, the tactile loses its ability quickly after circulation. Moreover, the existing mobile application specifically designed to help the visually impaired is usually unable to recognize Malaysian currency, especially coins. Additionally, most mobile applications require user subscriptions to enable full functionality, which limits the visually impaired users' ability to use them in daily life. Therefore, this project aims to develop a mobile application using Flutter for both Android and iOS platforms, with the transfer learning on a pre-trained YOLOv8 model to recognize and count new Malaysian banknotes and coins one by one in real-time. The app also utilized TensorFlow Lite to convert the model into a mobile-compatible format to run on mobile devices. Furthermore, the app designed with a user-friendly interface and accessibility features such as high-contrast text, audio feedback, and vibration notification. The mobile application can help visually impaired users recognize and count the banknotes and coins they are handling in a more accessible with mAP50 (mean average precision calculated at an intersection over union threshold of 50) up to 98.7%. A custom counting technique is also implemented, utilizing Euclidean distance calculations and a timeout mechanism for accurate object tracking
Job search self-efficacy, workplace anxiety and career exploration among final year students in Malaysia
Get PDFIncreasing technology changes the job market in Malaysia. Due to this, new job types are being created while old jobs are getting restructured. Therefore, final-year students must adapt to this by exploring careers through discovery of interests and research of different career options. In this regard, it is essential to review factors affecting career exploration behavior to properly address the issue. This study identifies Job Search Self-Efficacy (JSSE) and workplace anxiety as the factor that could influence career exploration. The main objective of the current study is to examine the relationships between JSSE, workplace anxiety and career exploration among final year students in Malaysia. Cross-sectional survey design was used, and data were collected via online questionnaires from 298 final-year students from Malaysia aged below 24 years by purposive sampling. A Spearman's rank correlation analysis indicated that there was a strong positive relationship between JSSE and career exploration, whereas the test results indicate that workplace anxiety did not significantly predict career exploration. Given that JSSE is a key predictor of career exploration, this study recommends that relevant authorities enhance career counseling to support students' career development. Additionally, recognizing the dual nature of workplace anxiety underscores the need for adaptive frameworks that address both its enabling and debilitating effects on career exploration. Future research should adopt mixed-methods and longitudinal designs to mitigate self-report biases and examine additional variables to gain a more comprehensive understanding of the factors influencing career exploration among final-year Malaysian students
Traffic sign detection from video for autonomous vehicles
Get PDFsign detection from video plays a vital role in enhancing the safety and decision-making capabilities of autonomous vehicles and Advanced Driver Assistance Systems (ADAS). This project focuses on the development of a robust deep learning-based detection system utilizing the latest YOLO11 model to identify and classify traffic signs from recorded video feeds. The system was trained using a carefully prepared dataset consisting of 21,688 images across 18 traffic sign classes, collected under various real-world conditions such as illumination changes and occlusions.
The YOLO11 model was fine-tuned through data augmentation and hyperparameter optimization to maximize detection accuracy and model generalization. The final model demonstrated strong performance, achieving a precision of 96.8%, recall of 97.3%, mAP@50 of 98.7%, and mAP@50–95 of 90.8%.
The project concludes with the successful implementation of an efficient and scalable traffic sign detection framework that supports high reliability. The findings contribute to the field of computer vision and intelligent transportation by demonstrating the effectiveness of the YOLO11 model in detecting traffic signs under challenging conditions. This work serves as a foundation for further enhancements in autonomous navigation and real-world deployment of intelligent perception systems