Browsing by Author "Khan, Rubayat Ahmed"

Now showing 1 - 17 of 17

A descriptive study on development of a transfer learning based fault detection model using 2D CNN for air compressors
(BRAC University, 2021-06) Hasan, Moenul; Uddin, A.K.M. Faiyaz; Ghosh, Anoup; Uddin, Jia; Khan, Rubayat Ahmed; Alam, Md. Golam Rabiul
Fault Detection is essential for the safe and efficient operation of industrial manufacturing. Successful detection of fault features allows us to maintain a stableproduc- tion line. Therefore, establishing a reliable and accurate fault detection method has become a huge priority now. Historically, various artificial intelligence-based models are used to predict faults in machines accurately to some extent. Mainly, machine learning and deep learning-based processes are being used. However, there are some shortcomings in those processes. Firstly, machine learning is mostly dependent on previous data and fails to recognize new issues that have not been introduced to the model during the training phase. Secondly, with deep learning, it is very time- consuming to reliably classify faults and difficult to establish an effective model for complex systems of current days. Thus, in our paper, we are proposing to use a transfer learning-based optimization of the deep learning process to meet the re- quirements ofreal-time fault classification and accurate detection of faultsin adverse operational conditions. We will be using wavelet transformation of raw signal data to 2D images and constructing a DCNN based transfer learning architecture toex- tract the fault features of the machine. Finally, we will be feeding the network with data from our target domain for fine-tuning the network to work accurately in the target domain. We will be testing with two cases to find the accuracy and accuracy optimization over the deep learning (AAG) values of our system. Finally, we will be comparing our architecture with state-of-the-art transfer learning architectures from Keras.
Analysis of financial data on the time series using data from the stock market
(BRAC University, 2022-05) Shachcha, Ifad Bhuiyan; Siam, Muhammad Ziaus; Rasel, Annajiat Alim; Khan, Rubayat Ahmed
Predicting financial data is really important for investors Often times investors do not have a proper tool to properly assess the market and forecast their predictions. Furthermore, not only investors in modern day civilians are also willing to invest as well and as there is an abundant amount of data available from the financial sector it is of utmost significance to find the optimal algorithm in a general case scenario. This project aims to show a comparison between the results found from some of the popular neural network algorithms. In this project we have employed the help of Dense Neural Network [DNN], Recurrent Neural Network [RNN], Long Short Term Memory unit [LSTM], Convolutional Neural Network [CNN] and a pipeline where we combined LSTM and CNN. We have kept some of the parameters similar and compared the results to determine an algorithm in a general case. This would help people take informed decisions while investing.
Automatic detection of defective rail anchors
(2014-11) Khan, Rubayat Ahmed; Islam, Samiul; Biswas, Rubel
Rail line anchors/fasteners are the metallic components that attach each line with the sleepers. These are essential rail components as absence of these often result in derailments. Therefore in order to prevent dangerous situations and ensuring safety rail lines are periodically inspected. Rail inspection in many countries especially in third world countries, like Bangladesh, is performed manually by a trained human operator who periodically walks along the track searching for visual anomalies. This manual inspection is lengthy, laborious and subjective. This paper presents a machine vision-based technique to automatically detect the presence of rail line anchors/fasteners using Shi - Tomasi and Harris - Stephen feature detection algorithms. This approach has confirmed to successfully detect scenarios with both grounded and missing anchors invoked in the experiment, with an accuracy of 83.55%, thus proving its robustness.
Automatic detection of defective rail anchors
(2014-11) Khan, Rubayat Ahmed; Islam, Samiul; Biswas, Rubel
Rail line anchors/fasteners are the metallic components that attach each line with the sleepers. These are essential rail components as absence of these often result in derailments. Therefore in order to prevent dangerous situations and ensuring safety rail lines are periodically inspected. Rail inspection in many countries especially in third world countries, like Bangladesh, is performed manually by a trained human operator who periodically walks along the track searching for visual anomalies. This manual inspection is lengthy, laborious and subjective. This paper presents a machine vision-based technique to automatically detect the presence of rail line anchors/fasteners using Shi - Tomasi and Harris - Stephen feature detection algorithms. This approach has confirmed to successfully detect scenarios with both grounded and missing anchors invoked in the experiment, with an accuracy of 83.55%, thus proving its robustness.
Automatic detection of defective rail anchors
(2014-11) Khan, Rubayat Ahmed; Islam, Samiul; Biswas, Rubel
Rail line anchors/fasteners are the metallic components that attach each line with the sleepers. These are essential rail components as absence of these often result in derailments. Therefore in order to prevent dangerous situations and ensuring safety rail lines are periodically inspected. Rail inspection in many countries especially in third world countries, like Bangladesh, is performed manually by a trained human operator who periodically walks along the track searching for visual anomalies. This manual inspection is lengthy, laborious and subjective. This paper presents a machine vision-based technique to automatically detect the presence of rail line anchors/fasteners using Shi - Tomasi and Harris - Stephen feature detection algorithms. This approach has confirmed to successfully detect scenarios with both grounded and missing anchors invoked in the experiment, with an accuracy of 83.55%, thus proving its robustness.
Automatic detection of defective rail anchors
(2014-11) Khan, Rubayat Ahmed; Islam, Samiul; Biswas, Rubel
Rail line anchors/fasteners are the metallic components that attach each line with the sleepers. These are essential rail components as absence of these often result in derailments. Therefore in order to prevent dangerous situations and ensuring safety rail lines are periodically inspected. Rail inspection in many countries especially in third world countries, like Bangladesh, is performed manually by a trained human operator who periodically walks along the track searching for visual anomalies. This manual inspection is lengthy, laborious and subjective. This paper presents a machine vision-based technique to automatically detect the presence of rail line anchors/fasteners using Shi - Tomasi and Harris - Stephen feature detection algorithms. This approach has confirmed to successfully detect scenarios with both grounded and missing anchors invoked in the experiment, with an accuracy of 83.55%, thus proving its robustness.
Automatic measurement of rail line expansion joint gaps
(© 2014 Institute of Electrical and Electronics Engineers Inc., 2015-03) Islam, Samiul; Khan, Rubayat Ahmed; Biswas, Rubel
Expansion joint gaps are the gaps which are deliberately left between the rail ends to allow for expansion of the rails in hot weather. Over gapping of these end to end gaps often result in derailments. Therefore in order to prevent dangerous situations and ensuring safety rail lines are periodically inspected. Rail inspection in many countries especially in third world countries, like Bangladesh, is performed manually by a trained human operator who periodically walks along the track searching for visual anomalies. Such manual inspection is lengthy, laborious and subjective. This paper presents a machine vision-based technique to automatically measure the length of rail line expansion joint gaps using morphological processing. This approach has confirmed to successfully detect scenarios of different condition with an accuracy of 89%, thus proving its robustness.
Automatic measurement of rail line expansion joint gaps
(© 2014 Institute of Electrical and Electronics Engineers Inc., 2015-03) Islam, Samiul; Khan, Rubayat Ahmed; Biswas, Rubel
Expansion joint gaps are the gaps which are deliberately left between the rail ends to allow for expansion of the rails in hot weather. Over gapping of these end to end gaps often result in derailments. Therefore in order to prevent dangerous situations and ensuring safety rail lines are periodically inspected. Rail inspection in many countries especially in third world countries, like Bangladesh, is performed manually by a trained human operator who periodically walks along the track searching for visual anomalies. Such manual inspection is lengthy, laborious and subjective. This paper presents a machine vision-based technique to automatically measure the length of rail line expansion joint gaps using morphological processing. This approach has confirmed to successfully detect scenarios of different condition with an accuracy of 89%, thus proving its robustness.
Automatic measurement of rail line expansion joint gaps
(© 2014 Institute of Electrical and Electronics Engineers Inc., 2015-03) Islam, Samiul; Khan, Rubayat Ahmed; Biswas, Rubel
Expansion joint gaps are the gaps which are deliberately left between the rail ends to allow for expansion of the rails in hot weather. Over gapping of these end to end gaps often result in derailments. Therefore in order to prevent dangerous situations and ensuring safety rail lines are periodically inspected. Rail inspection in many countries especially in third world countries, like Bangladesh, is performed manually by a trained human operator who periodically walks along the track searching for visual anomalies. Such manual inspection is lengthy, laborious and subjective. This paper presents a machine vision-based technique to automatically measure the length of rail line expansion joint gaps using morphological processing. This approach has confirmed to successfully detect scenarios of different condition with an accuracy of 89%, thus proving its robustness.
Automatic measurement of rail line expansion joint gaps
(© 2014 Institute of Electrical and Electronics Engineers Inc., 2015-03) Islam, Samiul; Khan, Rubayat Ahmed; Biswas, Rubel
Expansion joint gaps are the gaps which are deliberately left between the rail ends to allow for expansion of the rails in hot weather. Over gapping of these end to end gaps often result in derailments. Therefore in order to prevent dangerous situations and ensuring safety rail lines are periodically inspected. Rail inspection in many countries especially in third world countries, like Bangladesh, is performed manually by a trained human operator who periodically walks along the track searching for visual anomalies. Such manual inspection is lengthy, laborious and subjective. This paper presents a machine vision-based technique to automatically measure the length of rail line expansion joint gaps using morphological processing. This approach has confirmed to successfully detect scenarios of different condition with an accuracy of 89%, thus proving its robustness.
Clustering and detection of good and bad rail line anchors from images
(© 2015 Institute of Electrical and Electronics Engineers Inc., 2016-06) Islam, Samiul; Khan, Rubayat Ahmed
Absence of railway anchors/fasteners is a serious concern as it might lead to severe consequences such as train derailments. Hence regular inspection is an obligation to ensure safety. The third world countries choose the inspection process to be non-automatic where a trained operator moves along the rail line boarding a motor trolley checking for visual anomalies. In the previous research [1], an automatic system was proposed to overcome the cons of the running manual technique by using image processing. Two feature detection algorithms - Shi Tomasi and Harris Stephen - were used and an accuracy of 83.55% was achieved. This research presents an upgraded version of the previous work by introducing Neural Network. The addition of NN has not only speeded up the detection process but increased the accuracy significantly to approximately 93.86%.
Clustering and detection of good and bad rail line anchors from images
(© 2015 Institute of Electrical and Electronics Engineers Inc., 2016-06) Islam, Samiul; Khan, Rubayat Ahmed
Absence of railway anchors/fasteners is a serious concern as it might lead to severe consequences such as train derailments. Hence regular inspection is an obligation to ensure safety. The third world countries choose the inspection process to be non-automatic where a trained operator moves along the rail line boarding a motor trolley checking for visual anomalies. In the previous research [1], an automatic system was proposed to overcome the cons of the running manual technique by using image processing. Two feature detection algorithms - Shi Tomasi and Harris Stephen - were used and an accuracy of 83.55% was achieved. This research presents an upgraded version of the previous work by introducing Neural Network. The addition of NN has not only speeded up the detection process but increased the accuracy significantly to approximately 93.86%.
Clustering and detection of good and bad rail line anchors from images
(© 2015 Institute of Electrical and Electronics Engineers Inc., 2016-06) Islam, Samiul; Khan, Rubayat Ahmed
Absence of railway anchors/fasteners is a serious concern as it might lead to severe consequences such as train derailments. Hence regular inspection is an obligation to ensure safety. The third world countries choose the inspection process to be non-automatic where a trained operator moves along the rail line boarding a motor trolley checking for visual anomalies. In the previous research [1], an automatic system was proposed to overcome the cons of the running manual technique by using image processing. Two feature detection algorithms - Shi Tomasi and Harris Stephen - were used and an accuracy of 83.55% was achieved. This research presents an upgraded version of the previous work by introducing Neural Network. The addition of NN has not only speeded up the detection process but increased the accuracy significantly to approximately 93.86%.
Comparative analysis of machine learning techniques in optimal site selection
(BRAC University, 2023-01) Aurnab, Aukik; Choudhury, Shaktiman; Ruhan, Shoubhick Roy; Rifaiya Abrar, Shikh Muhammad; Hossain Rabbi, S.M. Riyadh; Khan, Rubayat Ahmed
Site selection is a crucial aspect of many businesses, as a company’s location can sig nificantly impact its success. In recent years, machine learning techniques have been increasingly used to assist with optimal site selection by providing data-driven pre dictions about the potential success of a given location. Machine learning techniques can be used to assist in the process of selecting the optimal site by analyzing the patterns in data such as demographics, lifestyle services, and geographic features. In this paper, we compare several machine learning techniques for their perfor mance in optimal site selection for features extracted from Open Street Map (OSM) data, WorldPop population data, and Bing satellite imagery. A target dataset cor responding to the features extracted was collected from Yelp data on restaurant check-ins, and this was used as a parameter to determine the human engagement rate of that location with the businesses in that area. Our analysis methods in cluded SVR, Random Forest, XGBoost, Ridge Regression, Lasso Regression, and ElasticNet. The satellite imagery collected from Bing maps were used to train CNN architectures such as; VGG16, VGG19, ResNet, DenseNet, and InceptionV3 and the results were compared. We evaluated the techniques using several metrics, including Root Mean Squared Error (RMSE), Mean Squared Error (MSE), Mean Absolute Error (MAE), Median Absolute Error(MedAE), Max Error(ME), and Median Abso lute Deviation(MAD). We used algorithm and strategies that performed the best in related works for this research. One meta model was also implemented in this work by an ensemble learning technique known as stacking. The model that performed the best for the data collected was then determined by looking at the error scores of different models. This work provides an insight into the strengths and limitations of each technique and recommendations for practitioners considering the use of ma chine learning in site selection. This study demonstrates the potential of machine learning for improving site selection processes and highlights the importance of con sidering multiple approaches.
Implementation of reinforcement learning architecture to augment an AI that can self-learn to play video games
(BRAC University, 2023-01) Mahmud, Aqil; Khan, Aswat Karim; Hasan Rafi, Mohammad Mehdi; Fahim, Kazi Rayhan; Rasel, Annajiat Alim; Khan, Rubayat Ahmed
This paper is intended to be a practical guide in terms of getting up and running with reinforcement learning. Ideally, it aims to bridge the gap between practi cal implementation and the theories available for RL. The theory of reinforcement learning involves two main components: an environment, which is the game itself and an agent, which performs an action based on its observation from the environ ment. Initially, no in-game rules will be given to the agent and it will be rewarded or punished based on the action that it will take. The goal is to increase Proximal Policy Optimization (PPO) to maximize the reward that our agent will get, so over time it will learn what action to take in order to do so. Therefore, we will develop an AI agent that will be able to learn how to play one of the most popular arcade games of all time, Street Fighter. We preprocess our game environment and apply hyperparameter tuning using PyTorch, Stable Baselines, and Optuna to do it. This approach will basically train different types of RL architecture and find a model with the most weighted parameters. Moreover, we are going to Fine Tune that model and run our test cases on it. We are going to see how a reinforcement learning algorithm learns to play.
Predicting crime using deep learning
(BRAC University, 12/24/2017) Shihab, Muhammad Nafees; Chowdhury, Anupam; Mahmood, SK. Belayet; Chakrabarty, Amitabha; Khan, Rubayat Ahmed
Criminal activities are available in every region of the world influencing social life and financial improvement. As such, it is a major concern of numerous legislatures who are utilizing distinctive advanced innovation to handle such issues. Crime Analysis, a sub branch of criminology, considers the behavioral example of criminal activities and tries to recognize the pointers of such events. Distinguishing the patterns of criminal activity of a place is vital in order to prevent it. Law enforcement organizations can work effectively and respond more rapidly if they have better knowledge about crime patterns in different geological points of a city. Deep learning agents work with data and utilize distinctive systems to discover patterns in data making it exceptionally helpful for predictive analysis. Law enforcement agencies utilize diverse patrolling techniques in light of the data they get the chance to keep a region secure. The aim of this paper is to use deep learning models to predict and classify a criminal incident by type, depending on its occurrence at a given location. The experimentation is conducted on a dataset containing crime records. For this supervised classification problem, we used a new approach - LSTM (Long Short Term Memory) and was able to classify crimes with 64.2% accuracy. CNN (Convolutional Neural Network) & Shallow dense model were used also. Solving the imbalanced class problem, the deep learning agent was able to classify crimes.
Real-time fire detection based on feature analysis using enhanced color segmentation and novel foreground extraction
(BRAC University, 2017-07) Khan, Rubayat Ahmed; Uddin, Jia
This research proposes two effective real time fire detection techniques, based on video processing. The former technique is restricted to indoor conditions only while the later does not have such constraints. Both the proposed methods utilize prominent features such as flame color information and spatiotemporal characteristics to identify fire areas. For the first technique, color segmentation is carried out in the earliest stage to separate potential fire areas using the red component of RGB. Moving pixels are identified using frame differences from a reference frame followed by de-noising. In the next phase of the model, the growth of the segmented regions of the current frame is compared with later frames and based on the fact that a hazardous fire expands with time, regions with no or decreasing growth is removed. The complex boundary of fire (rotundity) is valuable information that aids in detection. In the last step, a feature vector is created with rotundity information and trained using a neural network. The proposed model is tested using a dataset containing a wide range of indoor lighting conditions and compared to a state of the art fire detection technique to confirm its effectiveness. The experimental results show that the proposed model performs better compared to the state of the art model in terms of accurate detection and computation time, yielding an average accuracy of 99.1%.For the second technique, the initial stage of the work extracts fire colored pixels using a set of enhanced rules on RGB. Fire pixels are dynamic and to detect these moving pixels a novel method is proposed in this approach. The final verification is done by examining the area of the extracted regions. A harmful fire will grow over time, thus if the area happens to increase, the region under focus is declared as fire. Experimental results show that the model put forward outperforms other state of art models yielding an accuracy of 97.7%.