KGKapil Gupta

Kapil Gupta

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Kapil Gupta has a total of 28 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Gabor spectrogram-based deep learning framework for automated detection of diabetes mellitus

    by Kapil Gupta, Naresh Kumar, Pavan Chowdary Ravipati, Rahul Koshti, Jaideep Patel, Parbhat Gupta, Rovin Tiwari

  • Open Access

    Multi-model machine learning for automated identification of rice diseases using leaf image data

    by Rovin Tiwari, Jaideep Patel, Nikhat Raza Khan, Ajay Dadhich, Jay Kumar Jain, Kapil Gupta

    Rice, a staple meal for about half of the world’s population, is critical to global food security, especially in Asia. However, diseases have a severe impact on rice production, resulting in significant yield losses or outright crop failure. Traditional techniques of identifying rice diseases are time-consuming, labor-intensive, and rely heavily on specialist knowledge. As a result, a rapid, cost-effective, and automated method for detecting rice illnesses is critical for modernizing agricultural techniques and ensuring sustainable food production. This paper presents a novel hybrid deep-learning and machine-learning framework for automatically identifying rice plant diseases from leaf photos. We extracted deep features from rice leaf images using pre-trained CNN models—MobileNetV2, DarkNet19, and ResNet18. These features are then classified using machine learning classifiers with various kernel functions, which apply a strong 10-fold cross-validation technique to assure model reliability. Using a medium Gaussian kernel of the SVM classifier, the proposed system achieved a classification accuracy of 98.61%, specificity of 98.85%, and sensitivity of 97.25%. The framework is computationally efficient and scalable, allowing for greater dataset testing. The proposed technique provides a dependable and efficient solution for accurate identification of rice leaf diseases, reducing farmers’ reliance on manual inspection and supporting timely intervention.


  • Three-Dimensional Attractor Reconstruction for Enhanced ECG Classification of Arrhythmia and Congestive Heart Failure

    by Shresth Gupta, Manojeet Roy, Nalini Pusarla, Philip J. Aston, Kapil Gupta, Varun Bajaj

    This paper proposes a novel technique that leverages the entire ECG waveform to detect subtle shape changes in long-duration data streams. The approach employs symmetric projection attractor reconstruction to generate a 3-dimensional attractor from time-delayed segments of raw ECG waveforms. By projecting the attractor onto a two-dimensional plane, baseline variations are minimized, and novel quantitative features (radial density and theta density) characterizing the shape and variability of sensor data are extracted. These features are collected over sliding time windows and fused to serve as input for machine learning classifiers. To validate the proposed algorithm, a publicly available dataset comprising 162 ECG signals was utilized, including 96 cases of arrhythmia, 30 cases of congestive heart failure (CHF), and 36 cases of normal sinus rhythm (NSR). The proposed method achieved a validation accuracy of 94% and a test accuracy of 93.2% in classifying the three classes using a weighted K-nearest neighbour (KNN) model. Obtained results underscore the robustness and simplicity of the presented approach in diagnosing cardiac abnormalities, providing a promising pathway for enhanced biomedical sensor-based diagnosis.


  • CardioNet: A Lightweight Deep Learning Framework for Screening of Myocardial Infarction Using ECG Sensor Data

    by Kapil Gupta, Varun Bajaj, Irshad Ahmad Ansari

    yocardial infarction (MI) stands as one of the most critical cardiac complications, occurring when blood flow to the cardiovascular system is partially or completely blocked. Electrocardiography (ECG) is an invaluable tool for detecting diverse cardiac irregularities. Manual investigation of MI-induced ECG changes is tedious, laborious, and time-consuming. Nowadays, deep-learningbased algorithms are widely investigated to detect various cardiac abnormalities and enhance the performance of medical diagnostic systems. Therefore, this work presents a lightweight deep learning framework (CardioNet) for MI detection using ECG signals. To construct time-frequency (T-F) spectrograms, filtered ECG sensor data is subjected to the short-time Fourier transform, movable Gaussian window-based S-transform (ST), and smoothed pseudo-Wigner-Ville distribution methods. To develop an automated MI detection system, obtained spectrograms are fed to benchmark Squeeze-Net, Alex-Net, and a newly developed, lightweight deep learning model. The developed CardioNet with ST-based T-F images has obtained an average classification accuracy of 99.82%, a specificity of 99.57%, and a sensitivity of 99.97%. The proposed system, in combination with a cloud-based algorithm, is suitable for designing wearable to detect several cardiac diseases using other biological signals from the cardiovascular system.yocardial infarction (MI) stands as one of the most critical cardiac complications, occurring when blood flow to the cardiovascular system is partially or completely blocked. Electrocardiography (ECG) is an invaluable tool for detecting diverse cardiac irregularities. Manual investigation of MI-induced ECG changes is tedious, laborious, and time-consuming. Nowadays, deep-learning-based algorithms are widely investigated to detect various cardiac abnormalities and enhance the performance of medical diagnostic systems. Therefore, this work presents a lightweight deep learning framework (CardioNet) for MI detection using ECG signals. To construct time-frequency (T-F) spectrograms, filtered ECG sensor data is subjected to the short-time Fourier transform, movable Gaussian windowbased S-transform (ST), and smoothed pseudo-Wigner-Ville distribution methods. To develop an automated MI detection system, obtained spectrograms are fed to benchmark Squeeze-Net, Alex-Net, and a newly developed, lightweight deep learning model. The developed CardioNet with ST-based T-F images has obtained an average classification accuracy of 99.82%, a specificity of 99.57%, and a sensitivity of 99.97%. The proposed system, in combination with a cloud-based algorithm, is suitable for designing wearable to detect several cardiac diseases using other biological signals from the cardiovascular system.M.


  • Automated assessment of mental workload from PPG sensor data using cross-wavelet coherence and transfer learning

    by Shresth Gupta, Kapil Gupta, Anurag Singh

    Highly complex cognitive works require more brain power. The productivity of a person suffers due to this strain, which is sometimes referred to as a mental burden or psychological load. A person’s mental health and safety in high-stress working conditions can be improved with the help of mental workload assessment. A photoplethysmogram (PPG) signal is a non-invasive and easily acquired physiological signal that contains information related to blood volume changes in the micro-vascular bed of tissues and can indicate psychologically relevant information to assess a person’s mental workload (MW). An individual under a high MW possesses an increase in sympathetic nervous system activity, which results in morphological changes in the PPG waveform. In this work, a time-frequency analysis framework is developed to capture these distinguishing PPG features for the automatic assessment of MW. In particular, a cross-wavelet coherence (WTC) approach is proposed to extract simultaneous time-frequency information of the PPG during MW relative to the resting PPG. The suggested technique is validated on a publicly available data set of 22 healthy individuals who took part in an N-back task with PPG recording. Under three different fixed window lengths, images are obtained using WTC between PPG records during N-back task activity and rest. The images are used further to obtain PPG classification in two broad classes of low and high MW using a customized pre-trained Inception-V3 model. The best validation and test accuracy of 93.86% and 93.07%, respectively obtained in the window setting of 1200 samples used for WTC image creation.


  • Multiresolution Assessment of ECG Sensor Data for Sleep Apnea Detection Using Wide Neural Network

    by Kapil Gupta, Varun Bajaj, Saurabh Jain

    Sleep apnea (SA) is a non-communicable medical condition associated with sleep, characterized by recurrent interruptions in breathing during sleep. Electrocardiogram (ECG) sensor data are commonly employed in medical diagnosis to capture the minute electrical impulses of the heart and effectively diagnose SA complications. Analyzing raw ECG data to extract the underlying subtle features poses challenges due to its nonlinear and non-stationary nature. This article introduces a novel approach to diagnosing SA patients using a multi-resolution assessment of raw ECG data. The tunable Q -factor wavelet transform (TQWT) is used to decompose the raw signal into multiple sub-bands (SBs). Selecting the optimal basis parameters for accurate decomposition using TQWT is a complex task. To address this, an artificial gorilla troop optimization algorithm is integrated with the TQWT method to determine the best tuning parameters, resulting in more representative SBs. To differentiate between apnea and normal ECG segments, an array of statistical features is evaluated from all SBs. Selected features are then used as input for five different neural network configurations and four other machine learning modules. The wide neural network configuration achieves the highest accuracy of 96.60%, with a sensitivity of 96.50% and specificity of 96.41%. This research proposes a non-parameterized technique for the efficient decomposition of non-stationary data. The presented strategy, in conjunction with the neural network configurations, can also be applied to detect other health complications.


  • A Self-Assessment Framework for Initial Screening of Human Monkeypox

    by Kapil Gupta, Varun Bajaj

    Human monkeypox (MPX) is the deadliest infection that transmits from person to person by direct interaction with animals, bodily fluids, coughing or sneezing, and infected things such as bedding. Due to the symptoms of MPX infections being similar to measles and chickenpox, manual screening of MPX virus at an early stage is time-consuming, prone to community spread, and subject to human error. Computer-assisted detection of human MPX using skin lesion images may improve clinical efficiency and decrease the workload of health staff. The objective of this work is to develop a reliable system for self-diagnostic of MPX infection employing a cloud-based diagnosis system. In this paper, a publicly available MPX skin lesion database and a convolutional neural network (CNN) are used to detect MPX-infected subjects automatically. Two pretrained CNNs, and a newly-developed less-complex CNN, are used to diagnose MPX infection. Benchmarked models are trained using a transfer-learning strategy. A 10-fold holdout cross-validation scheme with a data augmentation strategy is employed to design a more robust network. Transfer learned DarkNet19 has attained a maximum classification accuracy of 96.74%. The efficiency of the presented work is demonstrated by the simulation results using the publicly accessible dataset. The experimental results demonstrate that CNN shows potential in the screening of MPX-infected subjects from skin lesion images. The presented framework can be an interface unit for cloud-based disease diagnostic systems.

  • INSOMNet: Automated insomnia detection using scalogram and deep neural networks with ECG signals

    by Kamlesh Kumar, Kapil Gupta, Manish Sharma, Varun Bajaj, U. Rajendra Acharya

    Sleep is a natural state of rest for the body and mind. It is essential for a human's physical and mental health because it helps the body restore itself. Insomnia is a sleep disorder that causes difficulty falling asleep or staying asleep and can lead to several health problems. Conventional sleep monitoring and insomnia detection systems are expensive, laborious, and time-consuming. This is the first study that integrates an electrocardiogram (ECG) scalogram with a convolutional neural network (CNN) to develop a model for the accurate measurement of the quality of sleep in identifying insomnia. Continuous wavelet transform has been employed to convert 1-D time-domain ECG signals into 2-D scalograms. Obtained scalograms are fed to AlexNet, MobileNetV2, VGG16, and newly developed CNN for automated detection of insomnia. The proposed INSOMNet system is validated on the cyclic alternating pattern (CAP) and sleep disorder research center (SDRC) datasets. Six performance measures, accuracy (ACC), false omission rate (FOR), sensitivity (SEN), false discovery rate (FDR), specificity (SPE), and threat score (TS), have been calculated to evaluate the developed model. Our developed system attained the classifications ACC of 98.91%, 98.68%, FOR of 1.5, 0.66, SEN of 98.94%, 99.31%, FDR of 0.80, 2.00, SPE of 98.87%, 98.08%, and TS 0.98, 0.97 on CAP and SDRC datasets, respectively. The developed model is less complex and more accurate than transfer-learning networks. The prototype is ready to be tested with a huge dataset from diverse centers.


  • Atrial Fibrillation Detection Using Electrocardiogram Signal Input to LMD and Ensemble Classifier

    by Kapil Gupta, Varun Bajaj, Irshad Ahmad Ansari

    Atrial fibrillation (AFib) is a type of heart arrhythmia, marked by an erratic and rapid contraction of the atria. Computer-aided diagnosis of AFib using electrocardiogram (ECG) sensor data may become a valuable tool in the detection and management of this common cardiac arrhythmia. In this letter, we present a new hybrid approach for the automatic classification of ECG signals using the local mean decomposition (LMD) and ensemble boosted trees classifier (EBTC). The LMD algorithm is employed to adaptively decompose the recorded ECG data into product functions (PFs). Four entropy-based features namely, log energy, sure, Shannon, and threshold entropy are computed from each PFs. The Kruskal-Wallis algorithm is employed to check the statistical significance of the obtained features and an EBTC is used for the screening of AFib episodes. The proposed technique achieved the highest classification accuracy of 92.33%, 90.33%, and 90.00% by classifying immediate terminating and non-terminating, terminates after one minute and non-terminating, immediate terminating and terminates after one minute AFib episodes, respectively. The presented method outperforms the existing machine learning-based approaches for detecting AFib using ECG data acquired from the publicly accessible AF Termination Challenge Database.