Notable Publications (Journal Papers)
1. MonkeyNet: A Robust Deep Convolutional Neural Network for Monkeypox Disease Detection and Classification
Authors: Diponkor Bala*, Md. Shamim Hossain*, Mohammad Alamgir Hossain, Md. Ibrahim Abdullah, Md. Mizanur Rahman, Balachandran Manavalan, Naijie Gu, Mohammad S. Islam and Zhangjin Huang;
Journal: [Neural Networks-Elsevier], February, 2023 [CiteScore: 14.5] [IF: 9.657; Q1]
The monkeypox virus poses a new pandemic threat while we are still recovering from COVID-19. Despite the fact that monkeypox is not as lethal and contagious as COVID-19, new patient cases are recorded every day. If preparations are not made, a global pandemic is likely. Deep learning (DL) techniques are now showing promise in medical imaging for figuring out what diseases a person has. The monkeypox virus-infected human skin and the region of the skin can be used to diagnose the monkeypox early because an image has been used to learn more about the disease. But there is still no reliable Monkeypox database that is available to the public that can be used to train and test DL models. As a result, it is essential to collect images of monkeypox patients. The “MSID” dataset, short form of “Monkeypox Skin Images Dataset”, which was developed for this research, is free to use and can be downloaded from the Mendeley Data database by anyone who wants to use it. DL models can be built and used with more confidence using the images in this dataset. These images come from a variety of open-source and online sources and can be used for research purposes without any restrictions. Furthermore, we proposed and evaluated a modified DenseNet-201 deep learning-based CNN model named MonkeyNet. Using the original and augmented datasets, this study suggested a deep convolutional neural network that was able to correctly identify monkeypox disease with an accuracy of 93.19% and 98.91% respectively. This implementation also shows the Grad-CAM which indicates the level of the model’s effectiveness and identifies the infected regions in each class image, which will help the clinicians. The proposed model will also help doctors make accurate early diagnoses of monkeypox disease and protect against the spread of the disease.
2. An Evaluation of Machine Learning Approaches for Early Diagnosis of Autism Spectrum Disorder
Authors: Rownak Ara Rasul, Promy Saha, Diponkor Bala*, S M Rakib Ul Karim, Md. Ibrahim Abdullah, Bishwajit Saha
Journal: [Healthcare Analytics-Elsevier], January 2024 [CiteScore: 11.2 ; Q1]
Autistic Spectrum Disorder (ASD) is a neurological disease characterized by difficulties with social interaction, communication, and repetitive activities. While its primary origin lies in genetics, early detection is crucial, and leveraging machine learning offers a promising avenue for a faster and more cost-effective diagnosis. This study employs diverse machine learning methods to identify crucial ASD traits, aiming to enhance and automate the diagnostic process. We study eight state-of-the-art classification models to determine their effectiveness in ASD detection.We evaluate the models using accuracy, precision, recall, specificity, F1-score, area under the curve (AUC), kappa, and log loss metrics to find the best classifier for these binary datasets. Among all the classification models, for the children dataset, the SVM and LR models achieve the highest accuracy of 100% and for the adult dataset, the LR model produces the highest accuracy of 97.14%. Our proposed ANN model provides the highest accuracy of 94.24% for the new combined dataset when hyperparameters are precisely tuned for each model. As almost all classification models achieve high accuracy which utilize true labels, we become interested in delving into five popular clustering algorithms to understand model behavior in scenarios without true labels.We calculate Normalized Mutual Information (NMI), Adjusted Rand Index (ARI), and Silhouette Coefficient (SC) metrics to select the best clustering models. Our evaluation finds that spectral clustering outperforms all other benchmarking clustering models in terms of NMI and ARI metrics while demonstrating comparability to the optimal SC achieved by k-means.
Publications (Journal Papers)
1. American Sign Language Alphabets Recognition using Convolutional Neural Network
Authors: Diponkor Bala*, Bappa Sarkar, Md. Ibrahim Abdullah and Mohammad Alamgir Hossain
Journal: International Journal of Knowledge Based Computer Systems, 2021
Sign language is a kind of language which is primary used for the deaf and hard of hearing and also used for those who are unable to physically speak. Sign language is only understood by a small percentage of the population. There is a huge communication gap between the deaf community and the hearing majority but it is not acceptable for a nation. Computerized sign language recognition capabilities attempt to break through this communication gap. Due to the advancement of technology, Artificial Intelligence has made human life easier by using various machine learning techniques. Computer vision is one of the most vital fields of artificial intelligence. However, utilizing computer vision to recognize American Sign Language is extremely difficult since sign language is extremely complicated and has a large inter-class variation. In the last few years, Convolutional Neural Network has become an effective way for the classification of multiclass images. In this paper, we have used Convolutional Neural Network (CNN) for the recognition of ASL alphabets. In this study, we have used the Sign Language MNIST dataset which consists of 34627 images where 27,455 training samples and 7172 testing data. The dataset includes 24 alphabets except for the letter J and Z among 26 alphabets. For the recognition of ASL alphabets, at first, we pre-processed our dataset by the normalization technique. After that, we designed a convolutional neural network architecture for extracting features from hand gesture images and then we trained our CNN model through the training dataset. We have finally evaluated our proposed CNN model based on the test dataset and obtained its accuracy to see that how many ASL alphabets recognize correctly. The proposed CNN architecture was able to achieve an accuracy of 99.78% on unseen data.
2. Childhood Pneumonia Recognition using Convolutional Neural Network from Chest X-ray Images
Authors: Diponkor Bala*
Journal: Journal of Electrical Engineering, Electronics, Control and Computer Science – JEEECCS, 2021.
In Bangladesh, Pneumonia is accountable for around 28% of mortality among children under the age of five. According to a recent study, one child dies from pneumonia every 39 seconds around the world. Pneumonia occurs because of microscopic organisms, infections or growths, and leaves youngsters battling for breath as their lungs load up with discharge and liquid. Delay in looking for suitable consideration and access to different sources for treatment are the fundamental hazard factors for pneumonia demise in small kids in Bangladesh. To fight this problem, this country needed an easily accessible quick solution. Due to the advancement of technology, Artificial Intelligence has made human life easier by using various machine learning techniques. In the last few years, Convolutional Neural Network has become an effective way for the classification of multiclass images. This paperwork has been developed a system in order to provide a simpler way to detect pneumonia in a short period of time. The model is developed with chest x-ray images taken from the frontal views to identify pneumonia. In this paper, Convolutional Neural Network (CNN) was utilized for the recognition of pneumonia disease. The CNN model is trained on a data set of 5,200 recently collected x-ray images. The dataset was divided into two classes, normal and pneumonia x-ray images. I have trained my model on different size x-ray images to evaluate its performance. This model can successfully detect pneumonia at an accuracy rate of approximately 98.87%. The proposed CNN model improves the pneumonia recognition accuracy than some existing methods.
3. Efficient Classification Techniques of Human Activities from Smartphone Sensor Data using Machine Learning Algorithms
Authors: Diponkor Bala*, G.M Waliullah
Journal: International Journal of Knowledge Based Computer Systems, 2020.
Increasing use of accelerometer and protractor sensors in recent years has created a field of study for the definition of human activities. This issue is tried to be solved by using machine learning methods. For this, it is solved by extracting different properties from the obtained signals, obtaining the characteristics specific to the activity and classifying these properties. In this study, the time and frequency domain properties of 4 different human activities were extracted, then a pre-treatment step was applied in accordance with the obtained feature set, and then the size was reduced with PCA and Fisher ‘LDA methods. The k-NN classifier and perceptron classifiers were designed for the obtained feature set and the classification process was performed. In this study, the classification success of these methods using different parameters has been examined and the results are shown.
4. Analysis of the Probability of Bit Error Performance on Different Digital Modulation Techniques over AWGN Channel Using MATLAB
Authors: Diponkor Bala*, G. M. Waliullah, Md. Nazrul Islam, Md. Ibrahim Abdullah and Mohammad Alamgir Hossain
Journal: International Journal of Knowledge Based Computer Systems, 2021.
Due to the demand at the present era of wireless communication technology, it is highly required a dependable communication system that can transmit more data with the lower probability of bit error. The digital modulation technique plays a vital role in modern wireless communication technology. Digital modulation technique provides the ability of more data transmission rates with better communication quality and higher data security using optimum bandwidth. By estimating the Probability of Bit Error, it will be possible to evaluate the quality of the performance of different modulation techniques. The aim of this paper is to discuss about the appropriate information of different digital modulation techniques which are extensively used in digital wireless communication systems. Finally by analyzing the Probability of Bit Error (BER) performance of various digital modulation techniques, it could be concluded that which modulation technique is suitable for different Signal-to-Noise Ratio conditions. This paper is especially focused on the comparison of the Probability of Bit Error (BER) performance among ASK, FSK, PSK and QAM modulation techniques. In this paper, all the simulation of ASK, FSK, PSK and QAM modulation techniques are accomplished by using MATLAB.
5. Analysis the Performance of OFDM Using BPSK, QPSK, 64-QAM, 128-QAM 256-QAM Modulation Techniques
Authors: Diponkor Bala*, Md. Shahabub Alam, Md. Nazrul Islam, Md. Ibrahim Abdullah and Mohammad Alamgir Hossain
Journal: International Journal of Knowledge Based Computer Systems, 2021.
At the present era of communication technology, every user wants a communication system that has higher data transmission capability and reliability. Orthogonal Frequency Division Multiplexing (OFDM) is a frequency division multiplexed multi carrier transmission method and each multiplexed signal is orthogonal to each other. This technology is known as core technology of the new generation wireless mobile communication systems. This technology has higher data transmission capability and reliability as well as has the ability to combat frequency selective fading or narrowband interference while maintaining high spectrum utilization. The aim of this paper is to analyze the performance of the OFDM technology using different modulation techniques. This paper is mainly focused on the calculation of the Bit Error Rate (BER) to analyze the performance of OFDM systems. In this paper, we considered BPSK, QPSK, 64-QAM, 128-QAM and 256-QAM modulation techniques. All the simulations are performed by using the MATLAB framework.
6. Study the BER Performance Comparison of MIMO Systems Using BPSK Modulation with ZF and MMSE Equalization
Authors: G. M. Waliullah, Diponkor Bala, Ashrafunnahar Hena, Md. Ibrahim Abdullah and Mohammad Alamgir Hossain*
Journal: Australian Journal of Engineering and Innovative Technology , 2020.
In the modern age, wireless communication is very helpful in various mobile antenna communication systems. In mobile communication systems, the transmission of data transfer rates is very high and it plays an important role in several services like video, top-quality audio, and mobile integrated service digital network. During the transmission of data at higher data transmission rates through the mobile radio channels, the channel impulse response can spread over many symbol periods as well as causes inter-symbol interference (ISI). Wireless transmission is suffering from fading and interference effects which may be combated with equalizer. Due to fading and interference, it creates a problem for signal recovery in wireless communication. The main objective of this paper is to analyze the different types of equalizers such as ZF and MMSE for BPSK modulation. The simulation result has been developed by using MATLAB toolbox version 2015a and a multi-tap ISI channel is considered. By analyzing the simulation result it shows that if the number of tap lengths is increasing, BER will decrease in ZF equalizer. And finally shows BER vs SNR comparison of two different types of an equalizer and is able to find out MMSE performance is better than ZF equalizer.
7. Simulation and Performance Analysis of OFDM System based on Non-Fading AWGN Channel
Authors: Diponkor Bala*, Md. Ibrahim Abdullah and Mohammad Alamgir Hossain
Journal: International Journal of Knowledge Based Computer Systems, 2022.
With the development of 4G network technology, gradually 5G wireless communication technology has also been derived and has been studied in deeply. 5G technology has been developed with based on 4G technology to strengthen its advantages, discard its shortcomings, and obtain further breakthroughs in functions. Due to the development of 4G technology, communication services such as downloading and transmitting large-volume data are being accomplished at an enormous speed. Orthogonal Frequency Division Multiplexing (OFDM) is a multi-carrier data transmission system that converts high-speed data streams into multiple parallel low-speed data streams by serial/parallel conversion, and then distributes them to sub channels on mutually orthogonal subcarriers of different frequencies for transmission. This technology has been recognized by the industry as the core technology of the new generation of wireless mobile communication systems. This paper mainly discusses the principle of OFDM-based LTE communication technology, and multi-channel simulation and analysis the performance of OFDM transmission system based on the MATLAB platform.
8. Analysis the Performance of OFDM-MIMO Channel with Different Equalizers
Authors: Diponkor Bala*, Md. Shamim Hossain, Md. Ibrahim Abdullah and Mohammad Alamgir Hossain
Journal: Journal of Network and Information Security, 2022.
The excellent efficiency, capacity, and dependability of today's wireless networks are concurrent to be achieved, and employing several communication methods antennas is an effective solution that has been extensively used. A communication system where both terminals are equipped with multi-antennas are referred to as MIMO systems, and when combined with OFDM technology are referred to as MIMO-OFDM. MIMO-OFDM has the ability to serve a large number of users with an enormous data transmission speed communication as well as utilizing the bandwidth efficiently. The aim of this simulation task explores three different equalization schemes in the MIMO flat fading channel, frequency-selective OFDM channel, and combined OFDM-MIMO wireless links on the bit error rate (BER) metric. Throughout the simulations, we modulate in 4-QAM (MIMO, OFDM-MIMO) and 16-QAM (OFDM) and observe BER performances for signal-to-noise ratio (SNR) up to 30. We find that given the specifications for OFDM as defined in IEEE 802.11a, precoding, and zero-forcing schemes in MIMO yield similar BER performances, while the MMSE scheme performs slightly worse at higher SNR's. Based on the equalization scheme, we assume perfect channel state information at the transmitter (CSIT) (for precoding) and the receiver (CSIR) (for zero-forcing and MMSE).
9. Design and Implementation of Low Cost Automatic Irrigation System using Microcontroller
Authors: Kowser Ahmed Shemul, Prianka Saha, Diponkor Bala*
Journal: Journal of Applied Information Science, 2022.
We are living in an era where we are facing a shortage in our water supply and other resources, primarily due to the ever-increasing population. The ever-increasing need to feed all this population is daunting. So there is a need to achieve maximum efficiency in utilizing these resources. Water is an important natural resource. The availability of freshwater is decreasing rapidly, that it is often said that if there is a third world war, it’d be for water. Usually, when we do agriculture, the most commonly faced challenge is water scarcity. Also, much of the water is not utilized by plants if we do the usual form of irrigation. Advanced irrigation techniques like automatic irrigation are starting to gain popularity these days. But they all need human intervention to monitor the process, which is tedious for large systems. We introduce our automatic irrigation system as a solution to this. We introduce a microcontroller-based automatic system, which actively senses the soil’s moisture level, opens a valve when water moisture is below a threshold and stops when the cutoff level has reached. Thus water wastage is minimized. This sensor can monitor a single plant or an entire area. A farmer can easily monitor this system. Most of the fields of agriculture will benefit from the proposed system of automatic irrigation.
10. Analysis the Performance of MIMO-OFDM for Various Modulation Techniques over AWGN, Rayleigh Fading and Rician Fading Channel
Authors: Diponkor Bala*, G. M. Waliullah, Md. Hafizur Rahman, Md. Ibrahim Abdullah and Mohammad Alamgir Hossain;
Journal: Journal of Network and Information Security, 2021.
Nowadays, we are living in the era of modern communication technology. The number of the mobile users is increasing tremendously day by day all over the world. Due to the increasing of the mobile users, the wireless communication systems are highly required a communication system that provides data transmissions rates and more reliability to the users. Multiple Input and Multiple Output Orthogonal Frequency Division Multiplexing (MIMO-OFDM) is being considered to get those facilities. MIMO-OFDM has the ability to serve a large number of users with an enormous data transmission speed communication as well as utilizing the bandwidth efficiently. The multicarrier modulation technique is cap-able of reducing the inter symbol interference and multipath fading problems. In this paper, we mainly focused on analysis the performance of MIMO-OFDM systems and the performance of MIMO-OFDM has been measured in terms of the Bit Error Rate and Signal to Noise Ratio based on different channels such as- AWGN, Rayleigh fading, Rician fading and different modulation techniques such as- BPSK, QPSK, M-PSK, D-BPSK, D-QPSK, DPSK and QAM. All the simulations are performed by MATLAB framework.
11. Design and Implementation of Low Cost Dual Axis Solar Tracking System using Microcontroller
Authors: Diponkor Bala*, G. M. Waliullah, Mohammad Alamgir Hossain;
Journal: Journal of VLSI Design and Signal Processing, 2021.
The phrase the sun is the source of all energy implies that solar energy is an essential element for the earth. Sun-powered vitality is fast becoming a substantial approach for renewable energy source assets. The sun is a plentiful source of vitality, and this sun-powered vitality may be effectively dealt with by employing sunlight-based photovoltaic cells and photovoltaic impact to convert sun-powered energy into electrical vitality. The solar tracking system maximizes the power generation of solar system by following the sun through panels throughout the day, optimizing the angle at which panels receive solar radiation. Compared to stable solar panels, a solar tracking system using solar panel linear actuators or gear motors can increase the efficiency of solar panels by 25% to 40%. The transformation efficiency of any sun-based application increases when the modules are consistently adjusted to the optimal edge as the sun crosses the sky. A dual-axis tracker allows panels to move on two axes, both north-south and east-west parallel. This paper presents the design and implementation of a dual-axis solar panel based on the Arduino microcontroller.
12. Performance Analysis of Zero Forcing and MMSE Equalizer on MIMO System in Wireless Channel
Authors: G. M. Waliullah, Diponkor Bala, Mst. Ashrafunnahar Hena, Md. Ibrahim Abdullah and Mohammad Alamgir Hossain*
Journal: Journal of Network and Information Security, 2020.
In wireless communication research multiple communication antennas are one of the major contexts. At present wireless communication is moving fast and the best example is MIMO. Wireless transmission is suffering from fading and interference effects which may be combated with equalizer. As a result of fading and interference, it creates a problem for signal recovery in wireless communication. The MIMO system uses Multiple Transmit and Multiple Receive antennas which take advantages of multipath propagation during a high distraction environment. This paper analyses the performance of Zero Forcing (ZF) and Minimum Mean Square Error (MMSE) equalizer for 2×2 and 4×4 MIMO wireless channels. By using MATLAB toolbox version 2015a simulation results can be got to the RF processing lab. The Bit Error Rate (BER) features for various communication antennas is simulated in the MATLAB toolbox and many merits and demerits of the system are discussed. The simulation results show that the equalizer based zero-forcing receiver is helpful for noise-free channel and is successful in removing ISI, but MMSE is an optimal choice than ZF in terms of BER characteristics.
13. Study the Performance of Capacity for SISO, SIMO, MISO and MIMO in Wireless Communication
Authors: Diponkor Bala, G. M. Waliullah, Mst. Ashrafunnahar Hena, Md. Ibrahim Abdullah and Mohammad Alamgir Hossain*
Journal: Journal of Network and Information Security, 2020.
Due to the rapid development of the wireless communication system, it is highly required a reliable system which can provide higher channel capacity and higher data transmission rates for the users. These are obtained by the Multiple Input Multiple Output (MIMO) systems because the MIMO systems allow the spatial diversity and spatial multiplexing technique due to its multiple antennas at both transmitter and receiver side. The aim of this paper is to discuss and show the capacity performance between SISO, SIMO, MISO and MIMO systems. In this paper, we will mainly be focused on the MIMO system due to its higher capacity and higher data transmission rates properties. For these properties of the MIMO systems, it will be perfectly suitable for modern communication technology.
14. Analysis the Performance of OFDM for Different Modulation Techniques with Channels and Image Transmission
Authors: G. M Waliullah, Nilufar Yasmin, Mohammad Alamgir, Ibrahim Abdullah, Diponkor Bala*
Journal: Journal of Electrical Engineering, Electronics, Control and Computer Science – JEEECCS, 2023.
Wireless applications' throughput, reliability, and bandwidth needs have grown in recent years. OFDM improves spectrum efficiency, data capacity, and fading resistance. OFDM is used in 3G GSM, WiMAX, and LTE. The OFDM system divides the input data stream into a number of lower rate streams that are sent over orthogonal subcarriers simultaneously to eliminate Inter Symbol Interference (ISI). The purpose of this paper is to evaluate the performance of the OFDM system utilizing various modulation and channeling techniques. Then, after analyzing the performance, we determined the optimal channel for an OFDM communication system. The final step is to transmit an image through the most optimal channel using various modulation techniques. Multiple channels, including AWGN, Rayleigh, and Rician, were utilized over an OFDM system, and M-PSK and M-QAM modulation techniques are used in both OFDM and image transmission. All simulation results are performed in the MATLAB toolbox 2019a.
15. Efficient Face Detection and Recognition with PCA and Eigenfaces: A Comprehensive Analysis
Authors: Md Ibrahim Abdullah*, Md Nazrul Islam, Diponkor Bala, Mohammad Alamgir Hossain, Md Atiqur Rahman, Md Shohidul Islam
Journal: International Journal of Advanced Networking and Application, 2025.
Face detection and recognition are critical tasks in computer vision with applications in security systems, biometric authentication, and human-computer interaction. This paper presents a comprehensive study leveraging Principal Component Analysis (PCA) and Eigenfaces for efficient dimensionality reduction and compact, discriminative facial feature representation. The study introduces a robust pipeline integrating preprocessing, feature extraction, and efficient training. Using the CelebA dataset for training and the LFW dataset for evaluation, the system addresses real-world challenges, including variations in lighting, expressions, and poses. The performance is analyzed across configurations, exploring the tradeoff between dimensionality reduction and recognition accuracy. Experimental results demonstrate that the PCA-based approach achieves high recognition accuracy while maintaining computational efficiency, making it suitable for resource-constrained environments. The findings highlight the system’s robustness, scalability, and practical applicability in both constrained and real-time scenarios. This work concludes with an analysis of strengths and limitations and offers recommendations for integrating non-linear techniques and advanced learning models to further enhance scalability, accuracy, and real-world performance.
Conference Proceedings (Published)
Authors: Diponkor Bala*, Mohammad Anwarul Islam, Mohammad Iqbal Hossain, Mohammed Mynuddin, Mohammad Alamgir Hossain and Md. Ibrahim Abdullah
Conference: 8th International Conference on Engineering and Emerging Technologies (ICEET) (Kuala Lumpur, Malaysia), 2022.
Recent advances in machine learning have employed deep learning to do several tasks. Deep learning has been used in the health sector to solve complex problems that require human intelligence. Without timely medical attention, the prognosis for patients with brain tumors is dismal. Radiologists are responsible for classifying tumors in radiographic images, which is a complex and time-consuming process that relies solely on their expertise. Modern radiology diagnosis, such as magnetic resonance (MR) scans, is largely subjective, putting patients at risk of damage. Use of Artificial Intelligence (AI) technology in order to avoid making mistakes when diagnosing is important to success. An automated approach for classifying different brain tumor classes in patients using magnetic resonance imaging (MRI) was suggested in this research, which focused on merging deep learning and radionics. We performed our work on three unique datasets with several classes. The proposed technique makes use of a convolutional neural network (CNN) as our deep learning model with the K-fold cross-validation concept in order to perform both binary and multiclass classification on our magnetic resonance imaging (MR) data. We took advantage of the power of CNN architecture in medical imaging. The model was trained and tested on random folded images from the dataset and was able to get an accuracy rate of 100%, 99.86%, and 100% in the corresponding dataset respectively, those are utterly remarkable, to put it mildly.
2. A Robust Plant Leaf Disease Recognition System Using Convolutional Neural Networks.
Authors: Diponkor Bala*, Mohammed Mynuddin, Mohammad Iqbal Hossain, Mohammad Anwarul Islam, Mohammad Alamgir Hossain and Md. Ibrahim Abdullah
Conference: 8th International Conference on Engineering and Emerging Technologies (ICEET) (Kuala Lumpur, Malaysia), 2022.
Plants are considered an energy supply to humanity. Plant diseases can damage farming, reducing harvest yields. This immediately affects farmers’ income and human health. Plant disease identification is one of the world’s most extensive challenges for farmers. Thus, leaf disease identification is vital in agriculture. Traditional disease detection approaches are difficult to detect in large numbers of plant leaf infectious illnesses. The ability to identify plant leaf diseases using images is rapidly improving. However, processing plant leaf images is difficult due to their complicated structure and shape. While modern deep learning algorithms can categorize and diagnose plant sickness, preparing plant leaf images is widely acknowledged as the most important and hardest step. Preprocessing has a big impact on the final results of deep learning. However, the proposed vision-based approaches efficiently detect and observe illness’s external aspects. We have used the New Plant Diseases Dataset in our paperwork. We proposed deep learning with a specially trained convolutional neural network (CNN), which can aid in the classification of plant leaf images. Our approach makes use of a CNN architecture that was trained on a collection of these plant leaf images. This method accurately classifies plant leaf images into 38 types of plant leaf diseases with 99.29% test accuracy, outperforming approaches previously defined as state-of-the-art.
3. Efficient Epileptic Seizure Recognition System using the Multi-model Ensemble Method from EEG
Authors: Diponkor Bala*, Mohammad Alamgir Hossain, Mohammad Anwarul Islam, Mohammed Mynuddin, Md. Shamim Hossain, and Md. Ibrahim Abdullah;
Conference: IEEE International Conference on Interdisciplinary Approaches in Technology and Management for Social Innovation (IATMSI) (ABV-IIITM, Gwalior, India), 2022.
DOI: 10.1109/IATMSI56455.2022.10119299 [Best Paper Award (2nd Place)]
As reported by the WHO, nearly 50 million people of all ages are affected by epilepsy globally. In the brain, epilepsy manifests as a condition that is marked by repeated, unpremeditated fits and other related problems, and epileptic seizures are hard to predict. The electroencephalogram (EEG) has evolved into an effective diagnostic tool for epilepsy and subsequent epileptic seizures. Visually analyzing an EEG chart helps determine seizures' duration and location. But it's time-consuming. To remedy this issue, we have developed a deep learning model-based framework. In this study, an EEG-based dataset is utilized to test and verify the binary classification tasks to predict and classify whether a patient is having an epileptic seizure. It is proposed to implement a deep learning-based multimodel ensemble model to be able to classify and identify EEG recordings showing or not epileptic seizures from EEG signals. The deep learning models considered are CNN, LSTM, and BiLSTM models, and the proposed ensemble model is composed of these three trained models with their best weights during training. The performance of these models is evaluated and compared by utilizing different metric analyses such as precision, recall, specificity, accuracy, and F1-score. The proposed model that we've made is the most accurate with 99.87% among all the models that have been experimented with in this research, and researching the different ways to solve the classification problem of epileptic seizures from EEG signals shows the best way to do it.
4. Effective Recognition System of American Sign Language Alphabets Using Machine Learning Classifiers, ANN and CNN
Authors: Diponkor Bala*, Mohammad Alamgir Hossain, Mohammed Mynuddin, Mohammad Anwarul Islam, Md. Shamim Hossain, and Md. Ibrahim Abdullah
Conference: IEEE International Conference on Interdisciplinary Approaches in Technology and Management for Social Innovation (IATMSI) (ABV-IIITM, Gwalior, India), 2022.
People who cannot talk are called audibly impaired, and they communicate with others through other means. The most popular method of communication is through sign language. American Sign Language (ASL) is the de-facto standard for sign languages taught globally. Automated sign language recognition tries to bridge the gap. Convolutional neural networks are the method of choice these days when it comes to the classification of multiclass images. To recognize ASL alphabets, we used a CNN, traditional machine learning classifiers, and an artificial neural network. The Sign Language MNIST dataset has a total of 34,627 image data, of which 27,455 and 7172 are training and test data respectively. Except for J and Z, the dataset comprises 24 alphabets. We used the training dataset to train our CNN, ANN, and other machine learning models. Then examined our proposed CNN model as well as other models including ANN on the test dataset to check how well they recognize ASL alphabets correctly. The traditional classifiers such as Linear Regression, Logistic Regression, Random Forest, SVM, and ANN were able to achieve an accuracy of 71.94%, 90.16%, 98.63%, 97.92%, 82.96% respectively whereas the proposed CNN model achieved 100 % of accuracy on the unseen test data.
5. SkinNet: An Improved Skin Cancer Classification System Using Convolutional Neural Network
Authors: Diponkor Bala*, Md. Ibrahim Abdullah, Mohammad Alamgir Hossain, Mohammad Anwarul Islam, Md. Atiqur Rahman, Md. Shamim Hossain
Conference: 4th International Conference on Sustainable Technologies for Industry 4.0 (STI 2022)) (Dhaka, Bangladesh), 2022.
According to current research, skin cancer is now considered to be among the most potentially lethal types of cancer that may occur in humans. Early detection of skin cancer, especially malignant type, can be tremendously advantageous as it may increase the survival rate of patients. Computers can help in the medical field by assisting with diagnosis. A convolutional neural network and image processing technology-based automated system were utilized in this study to recognise skin cancer. The system receives images of skin lesions, which are examined to determine the presence of skin cancer. The most prominent result of our computer based investigation is that it provides accurate results comparable with human analysis. In this research, we used the power of CNNs to bear on skin cancer recognition. We built a robust CNN model named SkinNet from scratch and trained it on the popular HAM10000 dataset. We then increased its performance using common techniques, such as the data balancing technique namely SMOTE, to address the issue of class imbalanced of data. We obtained vey good results, and we believe that in near future, CNNs will be able to outperform traditional diagnosis and probably replace expert dermatologists. Our proposed CNN architecture is capable of providing 98.60% of recognition accuracy on the data that has never been seen before. In fact, it will be enough to get significant results that could be used to enhance the survival rate of humans.
6. SERNet: A Novel Speech Emotion Recognition System Using Ensemble Deep Learning Approach
Authors: Diponkor Bala*, Mohammad Alamgir Hossain, Mohammad Anwarul Islam, Md. Atiqur Rahman, Md. Shamim Hossain and Md. Ibrahim Abdullah;
Conference: 4th International Conference on Sustainable Technologies for Industry 4.0 (STI 2022)) (Dhaka, Bangladesh), 2022.
Speech is among the most natural methods for us as human beings to express ourselves. Due to the relevance of emotions in today’s digital world of distant communication, their detection and analysis are crucial. Emotion recognition is extremely difficult since emotions are different for everyone. Speech emotion recognition is one area of application in which deep neural networks have excelled. Single learners have been used in the majority of the work done in this subject. We have developed a Speech Emotion Recognition (SER) system named SERNet that processes and classifies speech inputs to recognize emotions. As part of our research, we tried to explore emotional feelings in audio talks by looking at the acoustic characteristics of such recordings. This study proposes a novel approach using an ensemble of binary classifiers to simplify the multiclass classification problem into a binary classification problem, aiming to improve overall model performance. The binary classifiers are ensembled using a multilayer perceptron to obtain their final predictions on the multiclass classification problem. Utilizing a benchmark dataset designed specifically for the purpose of speech emotion recognition, the efficacy of this strategy has been proven. On the basis of the findings of the experiments, with an accuracy of 98.81%, this technique exceeds the most advanced models currently available.
7. A New Benchmark on Musculoskeletal Abnormality Recognition System using Deep Transfer Learning Model
Authors: Tajmin Sultana Mime, Diponkor Bala*, Mohammad Alamgir Hossain, Md. Atiqur Rahman, Md. Shamim Hossain and Md. Ibrahim Abdullah;
Conference: 3rd International Conference on Advancement in Electrical and Electronic Engineering (ICAEEE) (Gazipur, Bangladesh), 2024.
Musculoskeletal disorders are abnormalities of the bones, muscles, and joints that affect the great majority of people worldwide. Radiographic scans are the most widely used technique for detecting these aberrations as part of medical diagnostics. Early-stage detection of anomalies in radiographs is crucial for the patient. Moreover, bone fracture classification is costly, more time-consuming, and requires more effort. These reasons have made the deep learning-based classifier model a reliable alternative. In this study, we have applied a deep transfer learning technique to the problem of classifying such anomalies in radiographs of the musculoskeletal system. For multiclass recognition of radiological images into seven categories, the suggested model has been executed on the widely used MURA dataset, which includes 40,561 radiograph images. We utilized the DenseNet121 pre-trained model with some optimization in the suggested model. Our proposed model achieved the highest accuracy of 96.62%. The new model achieved a precision of 96.73%, a recall of 96.62%, an F1-score of 96.60%, and a Cohen Kappa score of 96.06%. We attain better performances that approximately 3% exceed the previous study and are also comparable to state-of-the-art performances.
8. ADSENet: Alzheimer’s disease classification using multiple deep residual learning with squeeze and excitation networks
Authors: SM Rakib Ul Karim, Diponkor Bala*, Rownak Ara Rasul, Sean Goggins;
Conference: 27th International Conference on Computer and Information Technology (ICCIT) (Cox's Bazar, Bangladesh), 2024.
Alzheimer’s disease is a degenerative neurological condition marked by the systematic deterioration and ultimate reduction in the size of nerve cells within the brain. Although it is highly challenging to cure, the most successful approach is to initiate therapy at an early stage of the illness. Timely identification of Alzheimer’s disease increases individuals’ likelihood of receiving effective therapy, making it crucial to obtain the condition promptly. Neuroimaging is a method used to classify Alzheimer’s disease based on the identification of structural abnormalities in the brain. This image-based categorization job is ideally suited to contemporary computer vision. The primary diagnostic approach is the evaluation of magnetic resonance imaging (MRI) images of the patient’s brain. We proposed a deep learning model that consists of multiple residual learning blocks with squeeze excitation networks, which can aid in early detection from brain MRI images with a high. Our approach uses a deep learning model architecture named ADSENet, which was trained on a collection of MRI images. The results indicate that volumetric measurements are highly effective in diagnosing Alzheimer’s disease with outstanding accuracy, so offering the possibility of early detection of the disease. This method achieves a test accuracy of 99.30% in the classification of MRI scans into four phases of Alzheimer’s disease, surpassing previously considered state-of-the-art approaches.
9. MRANet: A Modified Residual Attention Networks for Lung and Colon Cancer Classification
Authors: Diponkor Bala*, SM Rakib Ul Karim, Rownak Ara Rasul, Sraboni Ghosh Joya, Mohammad Alamgir Hossain, Zhaoxian Zhou;
Conference: 2nd International Conference on Next-Generation Computing, IoT and Machine Learning (NCIM) (Gazipur, Bangladesh), 2024.
Lung and colon cancers are the most common causes of cancer deaths. For treatment to work, it is very important to find out what is wrong as soon as possible and correctly. Learning models that use imaging technology to find different types of images have shown promise in automating the process of classifying cancer from histopathological images. The histopathological diagnosis is a big part of figuring out what kind of cancer someone has. This research aims to create a deep-learning model that can find lung and colon cancer in histopathological images with a high degree of accuracy. We suggested a new way to do things that is based on an enhanced version of the residual attention network structure. The model got its training from 25,000 high-resolution histopathological images that were split into groups. Our proposed model was achieved very high accuracy with scores of 99.30% for two classes, 96.63% for three classes, and 97.56% for five classes, respectively. These numbers are better compared to other modern architectures. This study shows that a deep learning model can accurately determine the difference between lung and colon cancer. Our proposed model’s better performance meets a key need in medical AI applications.
Upcoming Publication (Under Review)
Decalre soon..............
Authors:
Conference:
DOI:
Updated soon..........