Thesis (Bachelor of Science in Computer Science)

Browse

Now showing 1 - 20 of 885

3D Brain image segmentation using 3D tiled convolution neural networks
(BRAC University, 2023-09) Haque, Md Mahibul; Ria, Jobeda Khanam; Mannan, Fahad Al; Majumder, Sadman; Uddin, Md Reaz; Alam,Md. Ashraful
Gliomas are the primary brain tumors that are most commonly observed in adult patients and exhibit varying degrees of aggressiveness and prognosis. The accurate identification and diagnosis of Gliomas in surgical procedures heavily rely on the acquisition of precise segmentation results, which involve delineating the tumor region from magnetic resonance imaging (MRI) scans of the brain. The segmentation process in conventional 3D CNN methods is often reliant on patch processing as a result of the limitations in GPU memory. This paper presents an approach for segmenting brain tumors into distinct subregions, namely the WT, TC, and ET, utilizing a 3D tiled convolution-based segmentation method. The utilization of the 3DTC method enables the inclusion of larger patch sizes without requiring hardware with high GPU memory. This study presents three significant modifications to the standard 3D U-Net. Firstly, we incorporate 3D tiled convolution as the initial layer in our proposed models. Secondly, we substitute the trilinear upsampling layer with a dense upsampling convolution layer. Lastly, we replace the standard convolution block with recurrent residual blocks in the proposed R2AU-Net. The best framework was utilized to apply an average ensembling technique, aiming to achieve accurate results on the validation set of the BraTS 2020 dataset. The network proposed in this study was utilized for the analysis of the BraTS 2020 dataset. The evaluation of our method on the validation dataset yielded Dice scores of 90.76%, 83.39%, and 74.77% for the WT, TC, and ET regions, respectively.
3D character and overlay image movement by human body tracking using unity.
(BRAC University, 2020-04) Azim, Ashraful; Alam, Md. Golam Rabiul
Technology has made our life very easy and we have the touch of technology everywhere. By using technology, we are solving various difficult problems and making our life more convenient. In our daily life, we buy different clothing and ornaments from different shopping malls. In order to know if a dress is appropriate for the person properly, he or she needs to try the product before buying the product. But the process of trying a product in a big busy shopping mall is time-consuming and hectic for the customers. Often times, people find the trial room occupied and there are big queues in front of the trial rooms. So, we are planning to make the customer’s shopping experience hassle-free and convenient. For this reason, we come up with an idea of a smart mirror which will use overlay 3D image fitting techniques that fits overlay 3D images like dress, ornament or other user defined images within optimal space for various human body shape and type. For this, we are using human body tracking technology to first track the skeleton of the body and then take the coordination of the skeleton to move the 3D dress or object accordingly. So, the customer can select a dress and virtually try all the products without any hassle. This system will save the customer’s valuable time and they do not need to face the hassles of traditional trialing system. We also know that online shopping in growing its popularity day by day. By having this device at home, one can easily try dress they are about to buy from online. Before buying the dress he or she can easily check if the dress is appropriate for him or her by trying the product virtually without trying it physically. The shopkeepers too will be benefited from this product. During seasons like various religious vacations, shopkeepers are unable to provide enough trial room for the customers. Online shopkeeper can increase customer satisfaction as the customer can now try the product virtually from their residence. We believe that it is a great innovation that brings revolution in the customer’s shopping experience. This body tracking technology can be used in various other purposes also. 3D object movement is used in many fields. In gaming and film industry some games require tracking of human movement to move the 3D object accordingly.
6LoWPAN based smart home architecture for the smart future
(BRAC University, 2020-04) Hasan, Shah Muhammad Jannatul; Ali, Mohammad; Khan, Taufiqul Islam; Afrin, Farzana; Islam, Md. Motaharul
6LoWPAN is an emerging area of research. It helps to extend network architecture for WSN which allows end-to-end connectivity between a 6LoWPAN node and any IP devices of the network. To make a low power consumption mesh network, 6LoWPAN is a promising solution on a simple embedded system. Target is to carry IPv6 packets efficiently within a small link layer frame IEEE 802.15.4. Currently the absence of IP address hinders the process of accessing sensors directly. In remedy, IP supporting devices are necessary. Here according to 6LoWPAN technology each sensor will have a unique IPv6 address that will be auto-con gured. So, the user can easily access the sensors data directly from the Internet without any supporting device. In our home automation, we have implemented our proposal through simulator COOJA which is basically a simulator designed for wireless sensor network. It is a part of exceptionally versatile in performing various tasks and it works in framework Contiki. It has unlocked source where an integrated Internet Protocol Suite based on TCP/IP stack and multitasking takes only a few Kilobytes of RAM and ROM. A Contiki-based 6LoWPAN gateway is designed to interconnect 6LoWPAN nodes with the Internet IPv6. We have implemented our 6LoWPAN based network simulation in COOJA.We have found a better End to End communication, mobility, scalability and lower overhead which makes the system more convenient compare to other technology based home automation.
A 3D convolutional neural network architecture for early detection of coronary artery blockage (Coronary-3D-UNet)
(BRAC University, 2026-01) Sazid, Ahanaf Abid; Rahman, Mushfiqur; Akon, Md. Sabbir; Sadik, Jauad Ahmed; Chowdhury, Md. Jabed; Alam, Md. Ashraful
Coronary Artery Disease (CAD) is a major cause of death all over the globe where the estimated number of annual deaths are 9 million. Early and precise diagnosis of the stenosis of the coronary arteries is critical in providing clinical intervention and better patient outcomes. Despite the fact that invasive coronary angiography (ICA) is regarded as the most effective diagnostic tool, the procedure has certain risks and is not convenient enough to be used widely because of its inapplicability to whole population screening. Coronary Computed Tomography Angiography (CCTA) is a non-invasive, high-resolution alternative but manual analysis of CCTA images is time consuming and subject to inter-observer error. This paper presents Coronary-3D-Unet, a parameter-efficient 3D convolutional neural network to perform automated stenosis assessment and coronary artery segmentation on CCTA volumes. The residual learning and dual attention mechanisms (spatial and channel) and multiscale feature fusion are incorporated into the proposed architecture to improve vessel representation and resistance to various challenges such as small vessel structures, low contrast, and imaging artifacts. The framework allows automatic stenosis grading of severity as mild, moderate, and severe geometrically. On a publicly available benchmark dataset, experimental results demonstrate that Coronary-3D-Unet reaches a mean Dice Similarity Coefficient (DSC) of 76.6% and bests the official ImageCAS baseline with a significantly lower model complexity. The model is based on an efficient number of 3.8 million parameters, which is 83% less than the conventional dense 3D networks, so the model can be used to infer efficiently and be deployed practically. The model can also be integrated with clinical Picture Archiving and Communication Systems (PACS) that facilitate real-time analyses and better clinical usability.
A bayesian VAE based framework for synthetic data generation and false-alarm reduction in multi-class intrusion detection systems
(BRAC University, 2025-10) Bishal, M. Ridhwan Gani; Yeasin, Tasin Mohammad; Fuad, Mohammad Salah Akram; Rizvee, Raida; Mueed, Neamul; Hossain, Muhammad Iqbal; Humayun, Zayed
Intrusion detection systems (IDS) are constantly evolving in the field of network security to safeguard critical data assets against a growing array of sophisticated cyber threats, such as malevolent botnets, massive Distributed Denial of Service (DDoS) attacks, slow-rate DDoS attacks, advanced persistent threats (APTs), and zero-day exploits. Moreover, any organization’s network infrastructure remains vulnerable to different types of attacks, such as system abuse, security lapses, and break-ins. The Network Intrusion Detection System (NIDS) used in a network identifies such penetration attempts and intrusions. Researchers using deep learning (DL) have proposed increasingly capable IDS to protect critical networks; however, IDS are difficult to deploy in such environments because of high false-alarm rates (FAR). In this paper, we propose a hybrid framework that combines conditional variational autoencoder (CVAE)–based synthetic data generation with a Bayesian VAE model to reduce false-alarm rates in multi-class intrusion detection. This approach aims to lower FAR while maintaining strong detection performance by augmenting minority classes with class-consistent synthetic samples and leveraging calibrated Bayesian decisions.
A blockchain-driven framework designed for pharmaceutical community to secure and trace the trail of drug supply chain
(BRAC University, 2021-09) Rahman, Samiha; Aquib, Arif Awasaf; Jyoty, Watry Biswas; Rahman, Moshiur; Dewan, Tamanna; Hossain, Muhammad Iqbal
Our quality of life relies heavily on health products, which marks our dependence on the pharmaceutical industry. Drug safety is very important to ensure that our life-savers do not turn out to be the cause of our death. Due to the complicated and non-transparent supply-chain management that exists within this industry, we have to face the unfortunate risks of counterfeit drugs. Moreover, drug counterfeiters are taking advantage of people's vulnerabilities during COVID-19 and are making the situation even worse. A blockchain-based platform can make the process of drug de- velopment, production and marketing far more e cient and trackable compared to the existing system. Everything can be planned out, recorded and traced by a decen- tralized and distributed ledger technology. This will give end-consumers the control to monitor the products they receive. Our thesis paper aims to observe the issue with the existing supply-chain of pharmaceutical drugs and create a trustworthy infras- tructure which will e ectively make the overall process easier, while ensuring drug safety. We have used Hyperledger fabric, an open source enterprise-grade framework under the Hyperledger umbrella, to execute secure and well-planned transactions of drugs.
A classification and prediction based approach for real-time ETP outlet monitoring through E-IoT and remote sensing using machine learning and deep learning
(BRAC University, 2021-01) Hossain, Md. Mehedi; Mridha, Md. Jahid Hasan; Imran, Sazid Md.; Wahid, SK Ayub Al; Alam Md. Golam Rabiul
Water is a vital element in our environment but day by day water pollution is increasing in an alarming rate in our country. In Bangladesh’s perspective, industries such as textile and ready-made garments (RMG) contribute to a massive amount of waste or effluent. Effluent treatment plant (ETP) are used to remove as much suspended solids from wastewater as possible before it gets back to the environment. However, according to a report published by the Environment and forests ministry, seven state-run factories don’t have any effluent treatment plant (ETP) to treat their waste before disposal. And also even the factories which has ETP do not always keep the ETP up and running because it consumes a lot of electricity. The purpose of our research is to establish a setup which will monitor the real-time quality of water outside the industries and inform us whether the ETP is turned on or not with the help of E-IoT and various classification algorithm. It will also predict the seasonal impact where the ETP might be turned off again and what will be the quality of water with the help of various machine learning and deep learning algorithms such as CNN, KNN and LSTM. We have also tracking the sensor value for monitoring and the ETP outlet with RGB color analysis. We have successfully achieved an accuracy of 99% for KNN, 97.5% for CNN and 94.9% forecasting model accuracy for LSTM.
A coarse-to-fine hierarchical framework for bone marrow cell recognition: integrating morphological features with class-specific augmentation and multi-perspective explainable AI
(BRAC University, 2025-10) Mamun, Abdullah AL; Haider, Zarin Tasnim; Montaha, Sidratul; Jahan, Ismat; Mukta, Jannatun Noor; Nasim, Hamim Ibne
The classification of bone marrow (BM) cell is essential for the diagnosis of many haematological disorders. Automated cytological analysis still suffers from extreme class imbalance, very high morphological similarity between cell types, and poor interpretability of most artificial intelligence (AI) models, despite advances in medical imaging and deep learning. We present an interpretable, state-of-the-art framework for BM cell classification based on a large-scale dataset with 171,374 single-cell images annotated by experts with 21 classes. Given the high severity of the class imbalance (originally 3678:1 at times), we created a new subset of 95,865 images from the overall dataset through segmentation and feature extraction. Through this process, overrepresented classes were down-sampled, while specific types of augmentation were applied to underrepresented classes to restore balance, resulting in a ratio of 1435:1. Our recursive segmentation approach, based on CMYK (Cyan, Magenta, Yellow, and Black) and HLS (Hue, Saturation, and Lightness) colour spaces, reliably identifies nucleus, cytoplasm, and whole-cell regions. From these areas, we processed 144 biologically motivated shape, colour, texture, and fractal attributes. We build a hierarchical two-stage classification model named HierEff-S2, where an EfficientNet-B4 backbone assigns each cell to one of six morphological groups. Then, group-specific EfficientNet-B3 models perform fine-grained classification within each group. With 21 classes, this architecture obtains 86.1% accuracy and outperforms other models, including VisionMamba, Ensemble Model, and MobileNet. To promote clinical interpretability, we combine two explainable AI methods to visually highlight cell regions that lead to the model predictions: Grad-CAM and LIME. Using XAI, we report 95.2% correctness at the image level, thus providing biologically meaningful attention to the model.
A color vision approach based on the autoencoder technique and deep neural networks for reconstructing color images under various lighting conditions
(BRAC University, 2022-01) Faisal, Asm; Ahmad, Ashhab; Tazwar, Asif; Alam, Md. Ashraful
We present a color vision system that utilizes deep neural net- works to normalize pictures using the autoencoder algorithm. Image processing, encoding, and decoding are the three essen- tial processes in the proposed paradigm. An effective image processing approach is utilized to downsize acquired pictures into a finite image resolution equal to the number of input nodes of an autoencoder in the image processing section. En- coding and decoding procedures are included in the Autoen- coder. Second, a deep neural network-based encoding process creates a code for an input picture, and a deep neural network- based decoding process reconstructs the original image from the encoder’s code. Convolutional neural networks were used to train the autoencoder with over ten thousand scaled pic- ture datasets. The results of the experiments showed that the suggested model can recreate predetermined normalized pic- tures from original photographs, which may be employed in sophisticated color vision applications.
A color vision approach considering Reflection Co efficient based on Autoencoder techniques using deep neural networks
(BRAC University, 2021-09) Mahmud, Shakib Izaz; Shovon, Sartaz Islam; Hasnat, Md. Abrar; Na s, Md. Fahim; Alam, Md. Ashraful
Color vision approach using auto encoded technique is an effective way to detect objects. This approach considers various factors like movement detection, size and shape detection, color detection etc. Here we have considered reflection co efficient as another parameter to detect object material in different ambient lighting conditions. We are proposing to use deep learning methods to train our AI from values of light intensity of different objects in many controlled environments using digital illuminance meter also deep learning architecture on image data for detecting surface reflectance.
A color vision approach considering weather conditions based on auto encoder techniques using deep neural networks
(BRAC University, 2021-01) Raj, Mohammad Mainuddin; Tasdid, Samaul Haque; Nidra, Maliha Ahmed; Noor, Jobaer; Ria, Sanjana Amin; Alam, Md. Ashraful
Color vision approach is a riveting field of technology crucial in pioneering innovations like autonomous vehicles, autonomous drone deliveries, automated stores, robots, infrastructure and surveillance monitoring programs for security, manufacturing defect monitoring and more. When it comes to real life applications of automated machines, safety is a major concern and to ensure utmost safety the unpredictable has to be taken into consideration. We propose and demonstrate a color vision approach that allows image normalization hinged on autoencoder techniques employing deep neural networks. The model is composed of image preprocessing, encoding and decoding. The images are resized in preprocessing portion the images go through a cognitive operation where the input image becomes suitable to enter the autoencoding technique section. The autoencoder is comprised of two core components – encoder and decoder. To employ this system deep neural network is applied which generates a code of an image in the encoding process. Sequentially, the code changes over to decoding. Decoder portion decodes it and regenerates the initial image extracting it from the code of the encoder portion. It allows normalizing color images under different weather conditions such as images captured during rainy or foggy weather conditions. We devise it such that rainy and foggy images are normalized concurrently. The autoencoder is trained with numerous rainy and foggy datasets utilizing CNN. In this research, we investigate the model normalizing images in two different weather conditions – rainy and foggy conditions in real time. We used SSIM and PSNR to verify the accuracy of the model and confirm its capability reconstructing images in real time for advanced real life color vision implementations.
A color vision approach for reconstructing color images in different lighting conditions based on auto encoder technique using deep neural networks
(BRAC University, 2020-04) Gomes, Paul Richie; Uddin, Muhammad Arman; Sabuj, Hasibul Hasan; Faiz, Raian Ibn; Alam, Md. Ashraful
We propose a color vision approach that enables normalizing images based on autoencoder technique using deep neural networks. The proposed model consists of three main different steps: image processing, encoding and decoding. In the image processing part, an efficient image processing method is used to resize acquired images into a finite image resolution equal to the number of input nodes of an autoencoder. Autoencoder comprises encoding and decoding processes. Secondly, the encoding process based on deep neural networks generates a code of an input image and finally the decoding process using deep neural networks reconstructs the original image from the code generated by the encoder. The autoencoder is trained with more than ten thousand resized image dataset using convolutional neural networks. The experimental results verified that the proposed model enables reconstructing predefined normalized images from original images which can be used in sophisticated color vision applications.
A comparative analysis of emotion recognition using EEG signals with a channel selection technique
(BRAC University, 2021-01) Riduan, Jonayed Ahmed; Mahjabin, Most.; Mim, Nadia Tasnim; Islam, Ridwane-ul; Rana, Md. Shahriar Rahman; Parvez, Mohammad Zavid; Abrar, Mohammed Abid
Emotion can be defined as the neurophysiological changes people experience due to significant internal or external occasions. This is a mental condition that can a↵ect a person’s behavior, mood, way of life, and relationship with others. As it directly a↵ects one’s life, emotion recognition is an important subject in the area of research field. In recent years, there has been a relentless e↵ort to develop several models and datasets to detect human emotions and analyze them to understand the depth of complex human feelings and reduce error in the detection of emotions. To get better results in recognizing emotion, extensive research is needed to be done on the feature extraction methods and channel selection. While measuring the performance of di↵erent classification algorithms, it is very important to compare the results as well as preprocessing techniques. In this work, we extracted DWT wavelet features of the EEG channels from the DEAP dataset and used a statistical parameter Root Sum Square (RSS) to reduce the dimension of the features. Then we applied a channel selection algorithm on the preprocessed EEG data and selected ten channels with the highest average power. Finally, we classified positive and negative emotion related to valence and arousal using di↵erent classification algorithms (like KNN, RF, Bagging, Extra Tree, AdaBoost and MLP) for the selected EEG channels as well as for all EEG channels. The accuracy reports achieved for the selected channels were impressive; the highest test accuracy 67.58% for Valence was retrieved from the Bagging and Extra Trees classifier while MLP achieved the highest test accuracy result 63.67% for Arousal.
A comparative analysis of the different CNN-LSTM model caption generation of medical images
(BRAC University, 2023-05) Amin, Mahzabin Yasmin Binte; Shammo, Weney Hasan; Sayed, Jawad Bin; Hossain, MD Junaied; Alam, Md. Golam Rabiul; Reza, Md. Tanzim
The intent of this paper is to make the process of interpreting and understanding information within ultrasound pictures simpler and quicker by addressing the lack of techniques for automatically deciphering medical images. In order to do so, we propose a method of ultrasound image caption generation using AI that highlights the potential Machine Translation has in translating medical images to textual notations. The model needs to be trained on an ultrasound image dataset of the abdominal region including the uterus, myometrium, endometrium and cervix, a field of the medical sector that remains inadequately addressed. Two pre-trained CNN models, namely, VGG16 and Inception v3 have been used to extract features from the ultrasound images. Subsequently, the encoder-decoder model takes in two types of inputs, one for each of its layers. The two kinds of inputs are the text sequence and the image features. Both Vanilla LSTM and Bi-directional LSTM have been used to build the language generation model. The embedding layer along with the LSTM layer will process the text input. At last, the output from the two layers stated above will be merged.
A comparative analysis on solving university departmental course allocation problem using AI optimization algorithms
(BRAC University, 2021-01) Hasan, Shafqat; Dofadar, Dibyo Fabian; Khan, Riyo Hayat; Taj, Towshik Anam; Majumdar, Mahbubul Alam
This paper discusses about various types of constraints, regulations, difficulties and solutions to overcome the challenges regarding university departmental course allocation problem. A CSP solver algorithm, Genetic Algorithm, Simulated Annealing and a hybrid of Genetic Algorithm and Simulated Annealing has been used separately to generate the best course assignment and also to compare the results generated by these four algorithms. The Department of Computer Science and Engineering of BRAC University has been used as a case study to discover the scope of automation in this research. After analyzing the information gathered from the department itself, some constraints were formulated. These constraints manage to cover all the aspects needed to be kept in mind while preparing a class schedule for a faculty member without any clashes. The goal is to generate optimized solution(s) which will fulfill those constraints. At this point, the main focus is on the perspective of the faculty members but in the near future, there will be enough opportunities for expansions, like focusing on the lab change procedure of the students, assignment of student tutors and many more.
A comparative performance analysis of accident anticipation with deep learning extractors
(BRAC University, 9/29/2022) Mostak, Alfi Mashab; Neha, Nayna Jahan; Mohiuddin, Azwaad Labiba; Tabassum, Adiba; Hossain, Muhammad Iqbal; Abrar, Mohammed Abid
Accident anticipation has become a major focus to avert accidents or to minimize their impacts. Over the years, several network systems are being developed and applied in self-driving technology. Despite the fact that advancement in the autonomous industry is fast-growing, major efficiency is required in the network systems that are gradually emerging. Recent research has proposed a novel end-to-end dynamic spatial-temporal attention network (DSTA) by combining a Gated Recur- rent Unit (GRU) with spatial-temporal attention learning network, to identify an accident video in 4.87 seconds before the occurrence of the accident with 99.6% ac- curacy when tested on the Car Crash Dataset (CCD). However, DSTA has not been able to provide efficient results on the Dashcam Accident Dataset (DAD) dataset. Moreover, the GRU model integrated in the DSTA network has a weak information processing capability and low update efficiency amid several hidden layers. The decision-making process of the accident anticipation network may be understood using the high quality saliency maps produced by the Grad-CAM and XGradCAM approaches. In this paper, we evaluate that using Wide ResNet network enhances the performance mechanism of feature extraction to increase accident anticipation precision. This change improves the capacity to process information and the learning efficacy. In addition, we suggest employing a Gated Recurrent Unit (GRU) network which will serve as a prominent feature to train the model to recognize data’s sequential properties and apply patterns to forecast the following likely event. Hence, we plan to incorporate Wide ResNet50, a system for extracting features which will identify the vehicles at risk by using wider residual blocks. These neural networks generate labels for identifying hazardous conditions in driving environments in order to anticipate accidents.
A comparative study of AWS cloud native application deployment models
(BRAC University, 2020-10) Ahmed, Khandakar Razoan; Diganta, Sadifuzzaman; Islam, Md. Motaharul
Application deployment in cloud computing has become the most widely used system now. Developers use cloud computing instead of on-premises data centers in terms of deploying their application. For deploying our application, we need to follow some speci c models. For a long time, we have deployed our application in virtual machines. Nevertheless, nowadays the Docker container and Serverless models are gaining popularity due to its important features. Therefore, we need to know which model is better in terms of request-response, which model is error-prone and gives better throughput. In our paper, we have done a comparative study between these three native cloud application deployment models. The parameters of this study are HTTP response in a fixed time, error percentage, standard deviation, and throughput. We have used Amazon EC2 service for the Virtual Machine model, Amazon ECS service for the Docker Container model, and AWS Amplify for the Serverless model in the implementation part and Apache JMeter tool in the test part. In our study, we have found that the Serverless performed essentially better than the Docker container and Virtual Machine in the total HTTP response and throughput part although Docker Container did better than the Virtual Machine. And surprisingly, Virtual Machine did better than Docker Container and Serverless in the error percentage and standard deviation part.
A comparative study of deep learning methods for automating road condition characterization
(BRAC University, 2020-04) Ruhi, Zurana Mehrin; Sheetal, Farahatul Aziz; Prithu, Farisha Hossain; Arif, Hossain
Roads in Bangladesh provide infrastructural facilities to both agricultural as well as industrial sectors of the country. Distressed roads can cause fatal accidents as well as largely decelerate sector progress. This makes swift road inspection and repairs one of the most important aspects of our country’s holistic growth. As much as it aﬀects the general public, tackling this is as big a problem for the government as well. Currently, the problem for road repair is a multi-stage problem, which involves getting a complaint from a resident, physical road inspection by some oﬃcial, identifying the type of damage and then comes the process of actually repairing it. Here, we intend to make this cumbersome process simpler, by automating the problem identiﬁcation stage. We developed a method leveraging the Machine Learning and Deep Learning capabilities that can potentially detect a damaged road and identify the type of damage viz. pothole and crack. We self-captured data from the roads and streets, thus emulating the data we expect when this method is used in real-life by installing cameras on the city corporation’s garbage trucks. We reviewed various models ranging from conventional machine learning to complex deep learning algorithms and ultimately shortlisted three models: CNN, CNN-XGboost, and ResNet. These three models were then optimized for our problem, and then extensive testing was performed to determine the one that outperforms the rest. ResNet-34 emerged as a clear winner, with an accuracy of 87.8 % on the test data. Here, we’ll do an in-depth study of the eﬃcacy of these models on our problem statement.
A comparative study of lung cancer prediction using deep learning
(BRAC University, 2022-09) Mugdho, Aka Mohammad; Bhuiyan, Md. Jawad Hossain; Rafin, Tawsif Mustasin; Amit, Adib Muhammad; Chakrabarty, Amitabha; Rasel, Annajiat Alim
At the point when cells in the body develop out of control, this is alluded to as cancerous development. Lung cancer is the term used to depict cancer that starts in the lungs. At first in the field, classifier-based approaches are joined with various division calculations to utilize picture acknowledgment to recognize lung cancer nodules. This study found that CT scan images are more reasonable for delivering improved results than other imaging modalities. The use of the images is a piece of chiefly inspecting the CT scanned images that are viewed as informational collections for patients affected by lung cancer. The suggestion of our paper exclusively centers around the execution of concentrating on the calculation’s accuracy in diagnosing lung cancer. Thus, the primary plan of our examination is to utilize examined calculations to conclude which strategy is the most efficient method for detecting lung cancer initially. After training the model we found that Over all accuracy of Resnet-18 is 99.54%, the Overall accuracy of Vgg-19 is 96.35%, The overall accuracy of MobileNet V2 is 98.17%, Dense Net161 is 99.09% and Inception V3 is 98.17%. So we can see that ResNet18 perform better than other train model.
A comparative study of object detection models for Real Time Application in Surveillance Systems
(BRAC University, 2022-01) Alam, Saimun; Ahmed, Mahim Uddin; Hasan, Mehedi; Islam, Md. Morshedul; Arnob, Shahed Mehrab; Hossain, Dr. Muhammad Iqbal; Seraj, Mehnaz
In this paper, we attempted to give an overview based on thorough research and test ing of the latest object detection methods with an aim to help developers to build a Real Time Responsive CCTV Camera Model. As we welcome the 5G network worldwide, the coming future will surely be heavily dependent on smart machines and internet-based technologies. Therefore, we can assume that our daily life secu rity will also be managed by smart devices. In this research work, our aim is to do a thorough research on the latest models so that one can be chosen to implement and minimize the existing security system into a one device depended security system. The device we often use for surveillance and security purpose is CCTV camera. However, most of the cameras are not connected to the internet also they are not responsive. Which means, the outputs from the cameras cannot be used for further analysis by machines and can only be saved for manual check by humans. Our re search will help to develop such a system that will make the camera act like more of a security guard itself rather than a video recording device only. As we need to find out the best suited detection method we will check the accuracy, implementation process, power usage, GPU and CPU usage and then choose between previously invented methods such as HOG (Histogram of Oriented Gradients), Viola Jones De tector or the latest inventions such as R-CNN, SSD YOLO. Finally, this research will help the security device developers to choose the best algorithm and build cost efficient systems. Also, the future works of the research will help to create alert for abnormal presence of unknowns under surveillance automatically. Overall, we can say that our research will help to build more affordable, efficient and digitally secured home, offices, schools or any other buildings and even roads and highways in coming days.