eScholar-UMT
eScholar is the institutional repository for research conducted at UMT and maintains a large collection of theses, dissertations and projects produced by UMT graduates as part of their respective degree programs. It includes (but not limited to):
- PhD/MS Theses
- Graduate Program Research Projects
- Undergraduate Program Reports and Final Year Projects
- Full-text articles/research work of faculty and students
Join eScholar and sign-up to get the access of above mentioned digital resources at your desktop.
Communities in DSpace
Select a community to browse its collections.
Recent Submissions
Computer-aided detection of covid-19 using x-ray images
(UMT Lahore, 2022) Muhammad Ali; Rinsha Azaz; Aqsa Gul
Computer-Aided Detection using X-ray images provides an online way of solving the problems faced by almost every other person by saving their time and effort. The objective of this web app is to make the procedure of testing COVID-19 easier for every layman as well as an expert. This web app is a system for assessing the results and generating a report on its basis, whether the result comes positive or negative. The statistics section of the web app will display the number of users that have tried the web app and the results that they got, may it be positive or negative. The intention of building this project is to get the maximum accuracy in the test results. The algorithm used in the project is a Convolutional Neural Network having multiple layers. The model is
integrated with the help of a micro framework known as Flask and an html page is rendered to take an X-Ray image and display the results. Along with the results, the statistics will be updated in real time
An intelligent web application for predicting bone cancer
(UMT Lahore, 2022) Muhateer Muhammad; Muhammad Umair Khan
This thesis presents a study on the use of Efficient-Net, a pre-trained convolutional neural network (CNN) architecture, for the detection of bone tumors. Bone tumors are a serious health concern and early detection is crucial for successful treatment. The study focuses on the use of EfficientNetB5 model, which has been pre-trained on a large dataset of images, to detect bone tumors from X-ray images. The performance of the model was evaluated using a dataset of X-ray images from patients with known bone tumors. The results of the study show that the pre-trained Efficient-Net model achieved a high degree of 97% accuracy, sensitivity, and specificity in detecting bone tumors in X-ray images. The study also highlights the ability of pre-trained models like EfficientNetB5 to generalize well to different datasets, which can significantly reduce the time and resources required for training models from scratch. Overall, the study demonstrates the potential of Efficient-Net, a pre-trained CNN architecture, for the early detection of bone tumors, and provides a promising direction for future research.
Cargo shipping
(UMT Lahore, 2022) Sameer Saeed; FAISAL SARWAR; Muhammad HAMMAD SARWAR
We will provide users with a service that will help them to deliver their luggage, furniture, and machines through the cargo systems which are built already but do not provide a service to pick up from the sender. This will include tracking of the delivery.
We have integrated the chat into the service for cost estimation. This will save time for the customers and will ease their life a bit. Cargo shipping System provides you with the most innovative and best idea, which covers all aspects. We use React framework
with visual studio using node js. This is accomplished by utilizing the web. The cargo and delivery service is reliable. It describes methods of reliable calculation and evaluation of costs. It pays more attention to process operations and loading of the equipment. We provide an online cash system.
Identification of multiple skin diseases using convolutional neural networks
(UMT Lahore, 2022) Abdul Rafay
Skin conditions are widespread, impacting almost a third of people worldwide, yet their significance is often overlooked despite being easily visible. In reality, skin diseases are the fourth most common cause of all human illnesses. They are a significant global health concern, but diagnosing them can be difficult at times. Luckily, deep learning techniques have demonstrated promise in various areas, including identifying skin diseases. In this study, we blended two datasets to generate a new dataset of 31 skin diseases and used it to train the model. Three different architectures were tested for transfer learning on the skin disease dataset, which were EfficientNet, ResNet, and VGG. EfficientNet had the highest testing accuracy, and it was then fine-tuned further. The EfficientNet model was initially trained using 70% of the data, and it achieved a testing accuracy of 71%. However, it was later determined that the 70/30 data split was not effective, so the experiment was repeated with a 80/20 train-test split, resulting in an improved accuracy of 74%. Additional data augmentation was applied to improve the model's accuracy even more, resulting in a final accuracy of 86%. The model with the highest accuracy was then deployed as an API on a server and made available to the public to aid in early diagnosis and timely treatment of skin diseases.
Smart employee trackingface liveness and spoofing detection
(UMT Lahore, 2022) Muhammad Hassan Javed; Hamza Manzoor
Biometrics have emerged as a reliable solution for security systems, with facial recognition being one of the most heavily relied on and widely used biometric method for authorization. Face anti-spoofing technologies are critical in protecting face biometric
systems from spoofing attacks and malicious digital manipulation. Although high accuracies have been achieved from various studies, The problem of existing approaches' generalization remain uncertain and there still remains a significant gap for improvement as new emerging spoofing attacks, difficult illumination conditions and varying environments can really take a toll on the models' performance coupled with less diverse and less abundant data available on this domain. Addressing this problem, we propose a study by expanding the horizon of the studies done on face liveness and spoof detection by working on other color spaces such as LAB and HSV in order to potentially find more generalization features for identification of spoofs, and whether further studies should explore this dimension. The study is performed with one of the largest datasets available yet on face liveness and spoofing detection with such diverse sets of illumination conditions, environments and subjects. We have performed experiments in classification of spoofing in different color spaces and concluded that LAB and HSV is not up to the par in classification of spoofs. The other part of the study composed of training the model on combined dataset of LAB and HSV color space. The results from that model also could not beat the model trained on RGB color space. The RGB was the best performing model and the API was deployed using that model.