Ajay Namdeorao Paithane
@jspmrscoe.edu.in
Professor In E&TC
JAYWANT SHIKSHAN PRASARAK MANDALS RAJARSHI SHAHU COLLEGE OF ENGINEERING PUNE
EDUCATION
PhD in Electronics from PUNE University
RESEARCH INTERESTS
Embedded system, Biomedical Signal and Image Processing
FUTURE PROJECTS
Waste Management and Energy conservation
Applications Invited
Project in Health care
Scopus Publications
Scholar Citations
Scholar h-index
Scholar i10-index
Scopus Publications
R. Krishna Priya, Nitin N. Sakhare, Ajay Paithane, R. Shekhar, and M. Sabarimuthu
Wiley
AbstractA Stochastic Low‐Density Parity‐Check (LDPC) decoder is a type of 5G New Radio standard LDPC decoder that uses stochastic techniques to perform decoding. Stochastic LDPC decoding with 5G NR standard typically uses an iterative process, where messages exchanged among variable nodes (VN), check nodes multiple times. Stochastic LDPC decoders are often used in scenarios where the received signal is subject to varying levels of noise. They will provide improved error correction performance compared to traditional LDPC decoders, especially when dealing with channels with varying signal‐to‐noise ratios in 5G networks. Using the adaptive sparse quantization kernel least mean square algorithm (SLDPC‐ASQ‐KLMSA), this paper proposes an area‐efficient architecture design for a stochastic LDPC decoder. The LDPC code (2048, 1723) is taken from the LOGBASE‐T standard and used in this study. We examine the ASQ‐KLMSA connection effects. Starting with the VN. It makes checking node functioning easier and reduces inter‐connect complexity by capping extrinsic message length at 2 bits. Because of the simplified check node operation in ASQ‐KLMSA, the decoder nodes must exchange messages with a greater degree of accuracy. The 3–3 input grouping sub‐node of the degree‐6 VN was changed with an adder‐based 5–1 input grouping sub‐node for the (2048, 1723) code in order to get more accurate results when the check‐to‐variable messages aren't strong enough. A suggested decoder architecture was determined using a stochastic LDPC decoder developed for TSMC 65 nm process (2048, 1723). Bite error rate, throughput, mean square error, latency, power, and area usage are some of the metrics used to evaluate the effectiveness of the SLDPC‐ASQ‐KLMSA algorithm that has been suggested and implemented in Python. Thus, the proposed approach has attained 34.44%, and 38.39% low mean square error while compared with the existing methods such as higher‐performance stochastic LDPC decoder architecture designed through correlation analysis (HP‐SLDPC‐CA), Higher Throughput and Hardware Efficient Hybrid LDPC Decoder Utilizing Bit‐Serial Stochastic Updating(HLDPC‐BSSU), Flexible FPGA‐Based Stochastic Decoder for 5G LDPC codes (FPGA‐SD‐5G‐LDPC), respectively.
Sangita Ajit Patil and Ajay Namdeorao Paithane
Oriental Scientific Publishing Company
Stress affects mental and physical health, contributing to cardiovascular diseases and cognitive disorders, and early detection plays a crucial role in mitigating these risks. This study enhances stress detection by analyzing electroencephalography (EEG) signals from the DEAP ( A Database using Physiological Signals) data set and electrocardiogram (ECG) signals from the WESAD (Wearable Stress and Affect Detection) data set, with EEG offering a cost-effective solution and ECG providing detailed cardiovascular insights. It compares individual sensor analysis with multi-sensor fusion, demonstrating that fusion improves accuracy, as the ECG model achieves 91.79% accuracy, the EEG model reaches 96.6%, the feature-level fusion model achieves 98.6%, and the score-level fusion model achieves 97.8%. Using the Archimedes Optimization Algorithm (AoA) and Analytical Hierarchical Process (AHP) for feature selection and a hybrid Deep Convolutional Neural Network-Long Short-Term Memory (DCNN-LSTM) model for processing, the study highlights the effectiveness of a multi modal approach for real- time, accurate stress monitoring in clinical and industrial settings. It also integrates additional modalities and refines methods to enhance the system further, positioning AI-driven multimodal systems as powerful tools for early intervention and improved mental health management.
Amruta Jagadish Takawale and Ajay N. Paithane
National Taiwan University
Electroencephalogram (EEG) recordings typically capture the integrals of active brain potentials, which vary in latencies and populations. Anomalies in EEG data, often associated with epilepsy, play a crucial role in identifying conditions such as brain death, encephalopathy, coma, depth of anesthesia, and sleep disturbances. To get early warnings for these diseases, this work intends to propose a novel approach for brain activity detection from EEG signals. A new Hybrid Classification of Combined Coot Blue Monkey Optimization (HC-CCBO) method is proposed in this work. Initially, improved [Formula: see text]-score normalization was used to preprocess the EEG signal. Further, Discrete Wavelet Transform (DWT), improved correlation and statistical features were extracted. After that, we set up hybrid classification, which exploited Bi-Directional Gated Recurrent Unit (Bi-GRU) and Deep Max Out (DMO) models. Further, weights of BI-GRU and DMO were optimized via Combined Coot Blue Monkey Optimization (CCBO) optimization. Finally, we obtain the output scores of the rest, left fist, both fists, right fist, and both feet from the suggested hybrid brain activity recognition model. The effectiveness of the suggested HC-CCBO model is compared to conventional techniques using a variety of metrics. Compared to the existing models, the suggested model obtains a high maximum accuracy of 0.93 at the 90th learning percentage.
Dinesh B. Bhoyar, Swati K. Mohod, Rahul A. Burange, and Ajay Paithane
AIP Publishing
Amruta Jagadish Takawale and Ajay N. Paithane
Wiley
AbstractRecognizing brain activity from EEG waves is an important field of study in biomedical engineering and neuroscience. Conventional approaches usually begin with signal processing techniques to extract features from the EEG data, and then machine learning algorithms are applied to classify the data. However, the spatial resolution of these EEG signals is low, which makes it difficult to pinpoint the exact location of the neural activity source in the brain. There are ongoing initiatives to use DL‐based brain activity recognition algorithms to overcome these constraints. Therefore, this work presents a novel hybrid framework for brain activity detection using the enhanced Stockwell transform and an EEG signal that is called LinkNet and modified bidirectional–long short‐term memory (LN‐MBi‐LSTM) model. This framework follows a methodical approach that includes stages for feature extraction, brain activity recognition and preprocessing. Firstly, the improved Weiner filtering (IWF) approach is used to preprocess the EEG input signal. The relevant features are then extracted using a feature extraction technique from the preprocessed EEG signal. To identify the brain activity, these recovered feature sets are subsequently processed separately using LinkNet and modified bidirectional–long short‐term memory (MBi‐LSTM). A thorough analysis that takes into account both simulation and experimental calculations is part of the validation process for the LN‐MBi‐LSTM model. Finally, this study demonstrates the therapeutic potential of the LN‐MBi‐LSTM framework by presenting a strong and verified model for brain activity recognition. With the highest precision of 0.997, the LinkNet‐MBi‐LSTM model distinguishes itself from other models and confirms its exceptional capacity to produce accurate positive predictions.
Yogeshwari Nikam and Ajay N Paithane
IEEE
Heart disease poses a significant global challenge, being one of the leading causes of death worldwide. Recent advancements in machine learning (ML) applications have shown promising results in using electrocardiogram (ECG) data and patients' information to detect heart disease at an early stage. However, a prevalent issue is an imbalance found in the ECG data and patients' records, presenting a challenge for traditional ML methods to produce unbiased results. The explainability of the disease detection is lower in previous techniques. To address this, researchers and practitioners have explored various machine learning (ML) and Deep Learning (DL) methods over the years. This article systematically reviews the various ML and DL-based schemes utilized for heart disease detection. It focuses on the methodology, type of disease detection, merits, demerits, challenges and constraints of the research carried out for heart disease detection. It also provides the experimental analysis of the support vector machine (SVM, K-Nearest Neighbor Classifier (KNN), Naïve Bayes Classifier (NB), Classification Tree (CT), and Deep Convolution Neural Network (DCNN) for the heart disease detection on Framingham dataset. Further, the concise gaps that pave the way for future heart disease detection improvement are identified.
Sangita Ajit Patil and Ajay N. Paithane
IEEE
Electroencephalography (EEG) as a tool for analyzing mental stress has recently gained significant attention. The field of neuroscience employs electroencephalography (EEG) to investigate brain activity by measuring electrical signals on the scalp. The effective utilization of electroencephalography (EEG) for evaluating psychological stress has resulted in the emergence of innovative approaches for detecting levels of stress. This paper provides a comprehensive review of algorithms along with techniques used for mental stress analysis using EEG signals. The traditional algorithms discussed include power spectral density (PSD) analysis, Wavelet transform, and independent component analysis (ICA). This paper also explores the examination of more recent machine learning algorithms, including support vector machines, artificial neural networks, and fuzzy logic, for mental stress analysis. This paper examines the constraints of electroencephalogram (EEG) derived algorithms, prospects for further investigation, and a summary of the outcomes and their importance. Early investigations show that stress affects EEG signal power throughout frequency ranges. EEG signals may accurately distinguish stressed from non-stressed individuals using machine learning methods.
G. S. Mate, A. N. Paithane, and N. M. Ranjan
Springer Nature Singapore
Ajay N Paithane and Mukil Alagirisamy
Oriental Scientific Publishing Company
The human stress is a mental condition that can abnormally change the brain electrical activity, thus, electroencephalogram (EEG) signal measurements can detect and quantify those brain cognitive changes that are differentiated from the normal state. In this research work, EEG signals are used for the analysis and detection of the level of human stress. The EEG signals are collected from the human being called it as a subject under test. The stroop colour test has been used as a stressor to induce stress in the subjects under test. The various levels of stress in the stroop test have been verified to low, moderate, and high levels of stress in the subject. The input signals are then decomposed into the number of a narrowband signal using wavelet transform. During the experimentation important features are also extracted from EEG signal to identify normal and abnormal signals. The SVM classifier has been used in our research work for the classification of stress and non stress signals. The performance of the proposed system using SVM is comparatively good in dependent and independent systems. The highest accuracy achieved in this study is 90% (Standard Deviation = 0.015) for user-dependent systems and 72.3% (SD = 0.08) for user-independent systems. The results show that the proposed system is reliable for detecting stress and normal levels respectively.
Dnyaneshwar Vitthal Sahane, Pravin M Ghate, and Ajay N Paithane
IOP Publishing
Abstract Automation plays an important role in event welding domain. This paper describes the problems associated with arc tungsten inert gas (TIG) welding for automated welding in oil and gas industry for pipe joints. Welding joints inside the offshore pipelines is a complex process and rework or repetitive work is very costly. Weld imperfections are the foremost anxiety in the pipe joint welding. Return out the entire weld machine to find the defect at the time of welding. This is one of the popular industry come with the problem statement to develop the application specific solution for inner pipe welding quality welding. However, this paper gives the solution. Moreover, factors like Joint crack, moisture, corrugated weld object, excessive gas flow, Lack of fusion in the root, Dirty base or filler metal are major problems in the welding process. Due to these issues weld quality is impacted. There are several problem areas in the existing welding control system i.e., poor topology selection, less efficiency, high power loss, more size, high cost, more maintenance, less effective control algorithm for welding control etc Due to all these factors the welding quality is hampered in traditional welding system. Many researchers have worked on the power electronics hardware section improvement but very less work done on the software section. This paper discusses on the software algorithm development and how the weld quality improved. Software welding algorithm development is the novelty of this paper. This paper proposed and developed low-cost welding machine controller using STM32F103 microcontroller with protection scheme. A welding algorithm is proposed and experimented to enhance arc TIG welding quality and automation. Welding algorithm is tested & practically verified using phase-shifted full bridge (PSFB) power converter topology with pulse width modulation (PWM) control technology. The results reveal that the proposed algorithm significantly improves the quality of automated welding.
Inspired by the expansion of minimal effort advanced cameras in cell phones being conveyed in computerized systems, we think about the connection between perceptual picture quality and an exemplary PC vision errand of face recognition. We measure the corruption in execution of a well known and compelling face detector when human-saw image quality is corrupted by twists usually happening in catch, stockpiling, and transmission of facial pictures, including clamor, obscure, and pressure. It is observed that, inside a certain scope of picture quality, an unobtrusive increment in picture quality can radically enhance face recognition execution. These outcomes can be utilized to guide asset or transfer speed distribution in securing or correspondence/conveyance frameworks that are connected with face location undertakings. In this work a perceptual quality QualHOG feature is used. Face locators prepared on these new components give measurably huge change in resilience to picture bends over a solid gauge. Distortion dependent which is more distorted uninformed variations of the face indicators are proposed and assessed on a huge database of face pictures speaking to an extensive variety of mutilations. A one-sided variation of the preparing calculation is additionally recommended that further improves the power of these face locators. To encourage this exploration, we have developed another dataset in our lab for further study.
J.A. Mahajan and A. N. Paithane
IEEE
Here, evaluate the abasement in execution of well known and effective face detector when human captured picture quality is corrupted by additive gaussian noise and blur. It is observed that, inside a specific scope of recognized picture quality, an adequate increase in picture quality can improve face detection performance. These results can be utilized to guide data transfer capacity which regards with face detection task. A new face detector based on QualHOG features is proposed for robust face detection that increases image indicative Histogram of Oriented Gradients (HOG) features with perceptual quality-aware spatial Natural Scene Statistics (NSS) features. The new detector provides significant improvement in tolerance to image distortion. To improve this research, new face database containing face and non-face patches from pictures by variety of common distortion types and levels were created. Here we used 347 faces and 1287 non-faces database.
Mugdha Limaye and Ajay Paithane
IEEE
Nowadays there is an increased need to store images in all the fields such as medicine, engineering, industries. Mostly techniques like wavelet and discrete cosine transform have been implemented. Several techniques have been developed for lossy and lossless image compression, several techniques were developed. Image edges have limitations in capturing them if we make use 1-D wavelet transform simultaneously in 2 dimensions. This is because wavelet transform cannot effectively represent straight line discontinuities and be reconstructed in a proper manner like that of curvelet transform. The Curvelet Transform is more suitable for compressing images, which has more curved portions. Fast discrete curvelet transform is implemented is used that is implemented using stationary wavelet transform. The proposed method is tested on various medical images and the result shows better performance parameters like Peak Signal to Noise Ratio (PSNR) and Compression Ratio (CR) and Mean Square Error.
Reena Musale and A. N. Paithane
IEEE
Huge number of individuals are influenced by Diabetes Mellitus (DM) which is hard to cure because of its endless nature and hereditary connection. The uncontrolled diabetes may prompt heart related issues. Along these lines, the diagnosis and checking of diabetes is of extraordinary significance. By utilizing RR-interval signals, the automatic detection of diabetes can be performed. The nature of RR-interval signals is nonlinear and non-stationary. Subsequently linear strategies will be unable to catch the hidden data introduce in the signal. Empirical Mode Decomposition (EMD) is recently used in a broad range of applications for extracting signals from data generated in nonlinear and non-stationary processes. However, it has an Intrinsic Mode Function (IMF) consisting of signals of widely disparate scales. a new nonlinear method based on empirical mode decomposition (EMD) is proposed in this paper to discriminate between normal RR-interval signals and diabetic. By using IMFs of RR-interval signals the different parameters are extracted namely, the mean frequency parameter using Fourier -Bessel series expansion, amplitude modulation bandwidth, frequency modulation bandwidth and area computed from second order difference plot, Mean average value, First difference, ZCR. SVM (support vector machine) classifier has been utilized to measure the discrimination ability of the proposed features for detection of diabetic RR interval signals. The Results achieved from this proposed methodology indicate that these features provide the important difference between normal classes and diabetic. By using developed algorithm, the accuracy of analysis is 95%.
Pallavi Gaikwad and A. N. Paithane
IEEE
In this work, real time stress level recognition from Electroencephalogram (EEG) signal is proposed. Stress is a mental condition that effects the brain electrical activity to be different from the normal state. A system is designed to classify three levels of stress; Low Stress, Moderate Stress and High Stress. Validation of algorithm is carried out using by Stroop color-word test as stimuli to induce various levels of stress. EEG signals are collected using EEG reusable electrodes, which are placed on forehead scalp during the experiment. Measure of stress is taken by means of questionnaires method. Discrete Wavelet Transform (DWT) is used for pre-processing and using Support Vector Machine (SVM) as the classifier. Further analysis of stress is done by using trained algorithm as reference. Signals collected are continually given to algorithm to verify stress, three levels of stress can be recognized. Stress levels of the person under test are seen on screen using graph.
Anuja R. Bhagwat and A. N. Paithane
IEEE
The psychological disorders are generally appears in society. The prediction of such disorder is necessary in day to day life. Electroencephalogram (EEG) signal is neuronal activity of brain. A brain signal plays an important role for human disposition detection. EEG signals are non-linear in nature. In this paper, EEG signals are classified into four emotional states such as happy, angry, cry and sad. For development of such system the database which has been collected from the age group of 20 to 25 year old male and female both. In this system the wavelet transform used to measure statistical parameters. These parameters have been used to discriminate input signal based on the features, such as mean, mode, median, skewness, kurtosis etc. The Manhattan distance matrix concept of Hidden Markov Model has been used in this work to classify various dispositions. The Manhattan distance matrix compares trained dataset with input EEG signal. During experimentation on the EEG signal, it is observed that the various emotions like happy, angry, cry and sad are having unique and distinct feature set. The result indicates that signals bring more distinction between various dispositions from the EEG signals.
U. G. Patil, S. D. Shirbahadurkar, and A. N. Paithane
IEEE
Speech is natural vocalized and primary means of communication. Speech is easy, hand-free, fast and do not require any technical knowledge. Communicating with computer using speech is simple and comfortable way for human being. Speech recognition system made this possible. The acoustic and language model for this system are available but mostly in English language. In India there are so many peoples who can't understand or speak English. So the speech recognition system in English language is of no use for these people. Here we presented Isolated Hindi words recognition system which is a part of Automatic Speech Recognition (ASR) system. The main goal of ASR system understands a voice by computer or microphone and converts it into the text to perform required task. In this paper, we have used MFCC as feature extraction technique, Vector Quantization (VQ) with GMM (Gaussian Mixture Model) for recognition of Hindi isolated words. For practical analysis we prepared the Hindi words speech dataset of different males and females speakers.
Priyanka S. Ghare and A.N. Paithane
IEEE
Emotions plays very important role in person's daily life. Estimating human emotions through Electroencephalogram (EEG) has become one of the challenging area today. The representation of electrical activity of human brain is an important feature of EEG signal. The emotion detection using EEG signal follows the steps of emotion elicitation stimuli, i.e, collection or creation of the database, pre-processing, feature extraction, feature reduction or selection and classification. The combination of Adaptive Filter (Finite Impulse Response, FIR), Time-Frequency Analysis (Daubechies Wavelet Transform) and Support Vector Machine (SVM) classifier are used to detect the discrete emotions (happy, angry, sad and cry) of a human being using EEG signal.
U. G. Patil, S. D. Shirbahadurkar, and A. N. Paithane
IEEE
This paper presents a review on few notable speech recognition models that are reported in the last decade. Firstly, the models are categorized into sparse models, learning models and domain - specific models. Subsequently, the characteristics of the models have been observed using speech constraints, algorithmic constraints and performance constraints. The performance of these models reported in the literature is investigated and the findings are summarized. Eventually, the research gaps revealed by the literature are discussed and the need for Hindi based speech recognition system is substantiated.