Deep Learning Classification Model Research Articles

From deep learning to the discovery of promising VEGFR-2 inhibitors.

Vascular endothelial growth factor receptor 2 (VEGFR-2) stands as a prominent therapeutic target in oncology, playing a critical role in angiogenesis, tumor growth, and metastasis. FDA-approved VEGFR-2 inhibitors are associated with diverse side effects. Thus, finding novel and more effective inhibitors is of utmost importance. In this study, a deep learning (DL) classification model was first developed and then employed to select putative active VEGFR-2 inhibitors from an in-house chemical library including 187 druglike compounds. A pool of 18 promising candidates was shortlisted and screened against VEGFR-2 by using molecular docking. Finally, two compounds, RHE-334 and EA-11, were prioritized as promising VEGFR-2 inhibitors by employing PLATO, our target fishing and bioactivity prediction platform. Based on this rationale, we prepared RHE-334 and EA-11 and successfully tested their anti-proliferative potential against MCF-7 human breast cancer cells with IC50 values of 26.78±4.02 and 38.73±3.84 µM, respectively. Their toxicities were instead challenged against the WI-38. Interestingly, expression studies indicated that, in the presence of RHE-334, VEGFR-2 was equal to 0.52±0.03, thus comparable to imatinib equal to 0.63±0.03. In conclusion, this workflow based on theoretical and experimental approaches demonstrates effective in identifying VEGFR-2 inhibitors and can be easily adapted to other medicinal chemistry goals.

Factor annealing decoupling compositional training method for imbalanced hyperspectral image classification

Factor annealing decoupling compositional training method for imbalanced hyperspectral image classification

Deep-learning model for background parenchymal enhancement classification in contrast-enhanced mammography.

Breast Background Parenchymal Enhancement (BPE) is correlated with
the risk of breast cancer. BPE level is currently assessed by radiologists in Contrast-Enhanced
Mammography (CEM) using 4 classes: minimal, mild, moderate and marked, as described
in Breast Imaging Reporting and Data System (BI-RADS). However, BPE classification re-
mains subject to intra- and inter-reader variability. Fully automated methods to assess BPE
level have already been developed in breast Contrast-Enhanced MRI (CE-MRI) and have been
shown to provide accurate and repeatable BPE level classification. However, to our knowledge,
no BPE level classification tool is available in the literature for CEM.
Materials & Methods: A BPE level classification tool based on Deep Learning (DL) has
been trained and optimized on 7012 CEM image pairs (low-energy and recombined images)
and evaluated on a dataset of 1013 image pairs. The impact of image resolution, backbone
architecture and loss function were analyzed, as well as the influence of lesion presence and
type on BPE assessment. The evaluation of the model performance was conducted using dif-
ferent metrics including 4-class balanced accuracy and mean absolute error. The results of the
optimized model for a binary classification: minimal/mild versus moderate/marked, were also
investigated.
Results: The optimized model achieved a 4-class balanced accuracy of 71.5% (95% CI:
71.2-71.9) with 98.8% of classification errors between adjacent classes. For binary classifi-
cation, the accuracy reached 93.0%. A slight decrease in model accuracy is observed in the
presence of lesions, but it is not statistically significant, suggesting that our model is robust to
the presence of lesions in the image for a classification task. Visual assessment also confirms
that the model is more affected by non-mass enhancements than by mass-like enhancements.
Conclusion: The proposed BPE classification tool for CEM achieves similar results than
what is published in the literature for CE-MRI.

The effect of time normalization and biomechanical signal processing techniques of ground reaction force curves on deep-learning model performance

Time-series data are common in biomechanical studies. These data often undergo pre-processing steps such as time normalization or filtering prior to use in further analyses, including deep-learning classification. In this context, it remains unclear how these preprocessing steps affect deep-learning model performance. Thus, the aim of this study is to assess the effect of time-normalization and filtering on the performance of deep-learning classification models. We also investigated the effect of amplitude scaling. Using a public dataset (Gutenburg Gait Database, a ground reaction force database of level overground walking at self-selected walking speed involving 350 healthy individuals), we trained convolutional neural network (CNN) and long short-term memory (LSTM) models to predict binary sex (male, female) using three-dimensional ground-reaction forces to which we applied different processing approaches: zero padding, interpolation to 100% of signal, filtering, and scaling (min–max, body mass). The results show that transformations resulted in differences in model performances. Highest performance was obtained using unfiltered data, zero-padding, and min–max amplitude scaling (F1-score of 91 and 87% for CNN and LSTM, respectively). Not filtering data and using min–max scaling generally improve performance for both model architectures. For interpolation, results are not consistent across model architectures. This study suggests that processing steps must be considered in applications where deep-learning classification performance is relevant.

Cardiac Arrhythmia Classification Using Advanced Deep Learning Techniques on Digitized ECG Datasets.

ECG classification or heartbeat classification is an extremely valuable tool in cardiology. Deep learning-based techniques for the analysis of ECG signals assist human experts in the timely diagnosis of cardiac diseases and help save precious lives. This research aims at digitizing a dataset of images of ECG records into time series signals and then applying deep learning (DL) techniques on the digitized dataset. State-of-the-art DL techniques are proposed for the classification of the ECG signals into different cardiac classes. Multiple DL models, including a convolutional neural network (CNN), a long short-term memory (LSTM) network, and a self-supervised learning (SSL)-based model using autoencoders are explored and compared in this study. The models are trained on the dataset generated from ECG plots of patients from various healthcare institutes in Pakistan. First, the ECG images are digitized, segmenting the lead II heartbeats, and then the digitized signals are passed to the proposed deep learning models for classification. Among the different DL models used in this study, the proposed CNN model achieves the highest accuracy of ∼92%. The proposed model is highly accurate and provides fast inference for real-time and direct monitoring of ECG signals that are captured from the electrodes (sensors) placed on different parts of the body. Using the digitized form of ECG signals instead of images for the classification of cardiac arrhythmia allows cardiologists to utilize DL models directly on ECG signals from an ECG machine for the real-time and accurate monitoring of ECGs.

Research on lung sound classification model based on dual-channel CNN-LSTM algorithm

ulmonary diseases have a significant impact on human health and life safety, and abnormalities in the lungs are a direct response to lung diseases. Establishing an effective lung sound classification model that can assist in diagnosis is of great significance for electronic auscultation.In addressing the issue of lung sound signal classification, this study introduces a deep learning classification model based on a dual-channel CNN-LSTM algorithm. Initially, Mel-scale Frequency Cepstral Coefficients (MFCC) are employed for feature extraction from the dataset, transforming lung sound signals into Mel spectrograms. On this foundation, a dual-channel algorithm classification model is constructed, with parallel Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) modules. The CNN module is designed to capture spatial dimension features of the input data, while the LSTM module focuses on temporal dimension features. These two feature sets are fused together, enabling the model to classify lung sounds and thereby assisting in diagnosing pulmonary diseases for healthcare practitioners. This experiment used the ICBHI2017 Challenge Lungs dataset and obtained 5054 pieces of data through data augmentation and sampling techniques.The results show that the accuracy, recall, and F1 score of this model reach 99.01%, 99.13%, and 0.9915, respectively, significantly superior to other models, highlighting its practical application value.

Exploring non-Euclidean relationships in EEG emotion recognition: A dynamic graph attention network

Deep learning classification models based on electroencephalogram (EEG) emotion recognition have demonstrated considerable proficiency in the categorization of emotional states. However, these models have limitations in their capability to analyze the active states and cooperative relationships among distinct brain regions. This study proposes a dynamic graph attention network (DGAT) for EEG emotion recognition, which learns the features of each channel and leverages multiple-head self-attention mechanisms to capture non-Euclidean relationships between channels. Then, we use differential entropy features of emotions signals on the SJTU emotion EEG dataset (SEED). The DGAT model achieved improved subject-dependent and cross-subject classification accuracy compared to previous models. Moreover, ablation studies show that the channel weight matrix(CWM) and appropriate hyper-parameters can improve the performance of the DGAT model significantly. Furthermore, by conducting interpretable analysis of the new connections and electrode weights learned by the model, we find that these connection weight relationships reflect a certain degree of coordination within the brain for EEG-based emotion recognition. These findings provide a new method for EEG emotion recognition and highlights the potential for using deep learning models to analyze the active state and synergistic relationships among brain regions.

De novo antioxidant peptide design via machine learning and DFT studies

Antioxidant peptides (AOPs) are highly valued in food and pharmaceutical industries due to their significant role in human function. This study introduces a novel approach to identifying robust AOPs using a deep generative model based on sequence representation. Through filtration with a deep-learning classification model and subsequent clustering via the Butina cluster algorithm, twelve peptides (GP1–GP12) with potential antioxidant capacity were predicted. Density functional theory (DFT) calculations guided the selection of six peptides for synthesis and biological experiments. Molecular orbital representations revealed that the HOMO for these peptides is primarily localized on the indole segment, underscoring its pivotal role in antioxidant activity. All six synthesized peptides exhibited antioxidant activity in the DPPH assay, while the hydroxyl radical test showed suboptimal results. A hemolysis assay confirmed the non-hemolytic nature of the generated peptides. Additionally, an in silico investigation explored the potential inhibitory interaction between the peptides and the Keap1 protein. Analysis revealed that ligands GP3, GP4, and GP12 induced significant structural changes in proteins, affecting their stability and flexibility. These findings highlight the capability of machine learning approaches in generating novel antioxidant peptides.

Vision-based seam tracking for GMAW fillet welding based on keypoint detection deep learning model

Pre-programmed welding robots significantly improved the efficiency and quality of the welds in large batch production. In small and medium batch production, the robots need appropriate sensors to perform well and adapt to the changes and uncertainties in a noisy welding environment. Vision-based sensors enabled by machine learning are making it possible to sense in process previously not measurable. One challenge is developing artificial intelligent models capable of real-time seam tracking, particularly in fillet joints where visual analysis is hindered by non-perpendicular camera angles and arc reflections. In this paper, we propose a vision system that enables automated seam tracking with a collaborative robot. The vision-based deep learning classification model detects the tacks, where the seam is not visible. It is based on a keypoint detection deep learning model that addresses the challenges in distorted and noisy images of fillet joints between the pipes and flanges during the real-time Gas Metal Arc Welding to track the location of the seam in non tack images. The system is optimized for real time seam tracking by proposing the appropriate input image size. Multiple images and multiple points are also considered to provide a controllable signal of the location of the seam with less errors and outliers. Our proposed model can track the seam with more than 80 percent accuracy for errors less than 0.3 mm in fillet joints. The high accuracy of the proposed method would result in fewer flaws and defects and reduced rework, resulting in significant cost saving in manufacturing. The real-time monitoring also enables the adaptability to slight variations in gap.

Deep Learning Models for Classification of Lung Diseases

This thesis focuses on the importance of early detection in lung cancer through the use of medical imaging techniques and deep learning models. The current practice of examining nodules larger than 7 mm can delay detection and allow cancerous nodules to grow undetected. The project aims to detect nodules as small as 3 mm to improve the chances of early cancer identification. The use of constrained volume datasets and transfer learning techniques addresses the scarcity of medical data, and deep neural networks are employed for classification and segmentation tasks. Despite the limited dataset, the results demonstrate the effectiveness of the proposed models. Class activation maps and segmentation techniques enhance accuracy and provide insights into the most critical areas for diagnosis. This research contributes to the understanding of lung disease diagnosis and highlights the potential of deep learning in medical imaging.

Generalization of Deep-Learning Models for Classification of Local Distance Earthquakes and Explosions across Various Geologic Settings

Abstract Although accurately classifying signals from earthquakes and explosions at local distance (<250 km) remains an important task for seismic network operations, the growing volume of available seismic data presents a challenge for analysts using traditional source discrimination techniques. In recent years, deep-learning models have proven effective at discriminating between low-magnitude earthquakes and explosions measured at local distances, but it is not clear how well these models are capable of generalizing across different geological settings. To address the issue of generalization between regions, we train deep-learning models (convolutional neural networks [CNNs]) on time–frequency representations (scalograms) of three-component earthquake and explosion signals from eight different regions in the continental United States. We explore scenarios where models are trained on data from all regions, individual regions, or all but one region. We find that although CNN models trained on individual regions do not necessarily generalize well across different settings, models trained on multiple regions that include diverse path coverage generalize to new regions, with station-level accuracy of up to 90% or more for data sets from unseen regions. In general, CNN-based discrimination models significantly outperform models based on uncorrected P/S ratio (measured in the 10–18 Hz frequency band), even when CNN models are tested on data from entirely unseen regions.

AI-Based Network Intrusion Detection System

Abstract: This report discusses the research done on the chosen topic, which is Developing an AI-based Network Intrusion Detection System using ML and DL algorithms. Recently we have seen so much progress in Internet and communication technologies it is not just connecting computer networks and people but it is also connecting devices involving Big Data. It has so many benefits in each field which are crucial in today's world like education, health, digital transactions, traveling, and anything we can think of. With so many benefits it comes with its negative effects like cyber-attacks which can happen to anybody who is connected to the Internet. So, Networks and Security become very desirable areas of research and work. To be saved from the attacks we have to detect the cyber-attack and stop the intruder from causing harm to our system. We use IDS (Intrusion Detection System), so we can identify incoming attacks. A Network Intrusion detection system provides security by constantly monitoring the network traffic for suspicious behavior. For building an Intrusion Detection System we need a good dataset with a huge amount of data which is of good quality and can be used for training the System so it can predict the output more accurately. In this paper we used NSL- KDD [6] data set a refined version of the KDD’99 dataset. We developed many models using machine learning and deep learning and compared them for detecting intrusion in networks. We used supervised machine learning and deep learning techniques to train and build many classification models that can differentiate between attacking traffic and normal traffic. We compared the accuracy of every model of different datasets so we can find the model that is performing best for network intrusion detection. After performing all the research and comparison we found that a fully connected Deep Learning model is giving better performance than a machine learning model. We used Autoencoder for feature selection for the best-performing Deep Learning Classification Model.

AiGeneR 1.0: An Artificial Intelligence Technique for the Revelation of Informative and Antibiotic Resistant Genes in Escherichia coli.

There are several antibiotic resistance genes (ARG) for the Escherichia coli (E. coli) bacteria that cause urinary tract infections (UTI), and it is therefore important to identify these ARG. Artificial Intelligence (AI) has been used previously in the field of gene expression data, but never adopted for the detection and classification of bacterial ARG. We hypothesize, if the data is correctly conferred, right features are selected, and Deep Learning (DL) classification models are optimized, then (i) non-linear DL models would perform better than Machine Learning (ML) models, (ii) leads to higher accuracy, (iii) can identify the hub genes, and, (iv) can identify gene pathways accurately. We have therefore designed aiGeneR, the first of its kind system that uses DL-based models to identify ARG in E. coli in gene expression data. The aiGeneR consists of a tandem connection of quality control embedded with feature extraction and AI-based classification of ARG. We adopted a cross-validation approach to evaluate the performance of aiGeneR using accuracy, precision, recall, and F1-score. Further, we analyzed the effect of sample size ensuring generalization of models and compare against the power analysis. The aiGeneR was validated scientifically and biologically for hub genes and pathways. We benchmarked aiGeneR against two linear and two other non-linear AI models. The aiGeneR identifies tetM (an ARG) and showed an accuracy of 93% with area under the curve (AUC) of 0.99 (p < 0.05). The mean accuracy of non-linear models was 22% higher compared to linear models. We scientifically and biologically validated the aiGeneR. aiGeneR successfully detected the E. coli genes validating our four hypotheses.

An integrated deep-learning model for smart waste classification.

Efficient waste management is essential for human well-being and environmental health, as neglecting proper disposal practices can lead to financial losses and the depletion of natural resources. Given the rapid urbanization and population growth, developing an automated, innovative waste classification model becomes imperative. To address this need, our paper introduces a novel and robust solution - a smart waste classification model that leverages a hybrid deep learning model (Optimized DenseNet-121 + SVM) to categorize waste items using the TrashNet datasets. Our proposed approach uses the advanced deep learning model DenseNet-121, optimized for superior performance, to extract meaningful features from an expanded TrashNet dataset. These features are subsequently fed into a support vector machine (SVM) for precise classification. Employing data augmentation techniques further enhances classification accuracy while mitigating the risk of overfitting, especially when working with limited TrashNet data. The results of our experimental evaluation of this hybrid deep learning model are highly promising, with an impressive accuracy rate of 99.84%. This accuracy surpasses similar existing models, affirming the efficacy and potential of our approach to revolutionizing waste classification for a sustainable and cleaner future.

Identifying Diabetic Retinopathy in the Human Eye: A Hybrid Approach Based on a Computer-Aided Diagnosis System Combined with Deep Learning.

Diagnosing and screening for diabetic retinopathy is a well-known issue in the biomedical field. A component of computer-aided diagnosis that has advanced significantly over the past few years as a result of the development and effectiveness of deep learning is the use of medical imagery from a patient's eye to identify the damage caused to blood vessels. Issues with unbalanced datasets, incorrect annotations, a lack of sample images, and improper performance evaluation measures have negatively impacted the performance of deep learning models. Using three benchmark datasets of diabetic retinopathy, we conducted a detailed comparison study comparing various state-of-the-art approaches to address the effect caused by class imbalance, with precision scores of 93%, 89%, 81%, 76%, and 96%, respectively, for normal, mild, moderate, severe, and DR phases. The analyses of the hybrid modeling, including CNN analysis and SHAP model derivation results, are compared at the end of the paper, and ideal hybrid modeling strategies for deep learning classification models for automated DR detection are identified.

A deep learning-based theoretical protocol to identify potentially isoform-selective PI3Kα inhibitors.

Phosphoinositide 3-kinase alpha (PI3Kα) is one of the most frequently dysregulated kinases known for their pivotal role in many oncogenic diseases. While the side effects linked to existing drugs against PI3Kα-induced cancers provide an avenue for further research, the significant structural conservation among PI3Ks makes it extremely difficult to develop new isoform-selective PI3Kα inhibitors. Embracing this challenge, we herein designed a hybrid protocol by integrating machine learning (ML) with in silico drug-designing strategies. A deep learning classification model was developed and trained on the physicochemical descriptors data of known PI3Kα inhibitors and used as a screening filter for a database of small molecules. This approach led us to the prediction of 662 compounds showcasing appropriate features to be considered as PI3Kα inhibitors. Subsequently, a multiphase molecular docking was applied to further characterize the predicted hits in terms of their binding affinities and binding modes in the targeted cavity of the PI3Kα. As a result, a total of 12 compounds were identified whereas the best poses highlighted the efficiency of these ligands in maintaining interactions with the crucial residues of the protein to be targeted for the inhibition of associated activity. Notably, potential activity of compound 12 in counteracting PI3Kα function was found in a previous in vitro study. Following the drug-likeness and pharmacokinetic characterizations, six compounds (compounds 1, 2, 3, 6, 7, and 11) with suitable ADME-T profiles and promising bioavailability were selected. The mechanistic studies in dynamic mode further endorsed the potential of identified hits in blocking the ATP-binding site of the receptor with higher binding affinities than the native inhibitor, alpelisib (BYL-719), particularly the compounds 1, 2, and 11. These outcomes support the reliability of the developed classification model and the devised computational strategy for identifying new isoform-selective drug candidates for PI3Kα inhibition.

Classification of Diagnostic Certainty in Radiology Reports with Deep Learning.

A radiology report is prepared for communicating clinical information about observed abnormal structures and clinically important findings with referring clinicians. However, such observations and findings are often accompanied by ambiguous expressions, which can prevent clinicians from accurately interpreting the content of reports. To systematically assess the degree of diagnostic certainty for each observation and finding in a report, we defined an ordinal scale comprising five classes: definite, likely, may represent, unlikely, and denial. Furthermore, we applied a deep learning classification model to determine its applicability to in-house radiology reports. We trained and evaluated the model using 540 in-house chest computed tomography reports. The deep learning model achieved a micro F1-score of 97.61%, which indicated that our ordinal scale was suitable for measuring the diagnostic certainty of observations and findings in a report.

Mapping of groundwater availability in dry areas of rural and urban regions in India using IOT assisted deep learning classification model

Groundwater is a crucial resource for fulfilling the water requirements of India's rural and urban areas. The heterogeneous nature of geological, hydrological, and climatic factors results in substantial variability in the accessibility of groundwater across disparate regions. The present investigation centers on the cartography of groundwater accessibility in arid zones of rural and urban Indian areas using a Deep Learning Classification Model (DL-GWCM) supported by the Internet of Things (IoT). The introductory section underscores the importance of groundwater in India, where groundwater sources cater to around 80% of rural and 50% of urban water demands. The text highlights statistical data derived from surveys that indicate a notable decrease in groundwater levels. This underscores the pressing necessity for implementing effective monitoring and management strategies. The DL-GWCM is a proposed solution that aims to enhance the precision and effectiveness of groundwater availability mapping by incorporating IoT technology and Deep Learning Classification. The DL-GWCM comprises multiple constituent elements, such as Groundwater Prediction, Water Quality Index, and Conventional Neural Network- Bidirectional Long Short-Term Memory (CNN–Bi LSTM) classification. The process of Groundwater Prediction involves the utilization of past data and environmental factors to make precise forecasts of groundwater levels. The Water Quality Index evaluates the quality of subsurface water resources, guaranteeing their secure and enduring utilization. The Deep Learning Classification Model with IoT technology was implemented for groundwater accessibility mapping in Indian arid zones. It integrates Groundwater Prediction, Water Quality Index, and CNN–Bi LSTM classification. The model makes precise forecasts using past data and environmental factors, ensuring secure water quality. Using the CNN–Bi LSTM classification model improves the precision of groundwater availability mapping due to its resilient classification capabilities. These findings suggest that the DL-GWCM outperforms conventional approaches. The mean values of all five metrics for the proposed method are presented as follows: The performance metrics of the model are as follows: Root Mean Square Error (RMSE) of 0.77%, Mean Absolute Error (MAE) of 2.13%, Relative Absolute Error (RAE) of 8.72%, Root Relative Squared Error (RRSE) of 0.92%, and Correlation Coefficient (CC) of 0.92. The results of the proposed methodology facilitate the discernment of regions with abundant or scarce groundwater accessibility, thereby supporting the sustainable management and planning of groundwater resources.

Federated learning model for credit card fraud detection with data balancing techniques

In recent years, credit card transaction fraud has resulted in massive losses for both consumers and banks. Subsequently, both cardholders and banks need a strong fraud detection system to reduce cardholder losses. Credit card fraud detection (CCFD) is an important method of fraud prevention. However, there are many challenges in developing an ideal fraud detection system for banks. First off, due to data security and privacy concerns, various banks and other financial institutions are typically not permitted to exchange their transaction datasets. These issues make traditional systems find it difficult to learn and detect fraud depictions. Therefore, this paper proposes federated learning for CCFD over different frameworks (TensorFlow federated, PyTorch). Second, there is a significant imbalance in credit card transactions across all banks, with a small percentage of fraudulent transactions outweighing the majority of valid ones. In order to demonstrate the urgent need for a comprehensive investigation of class imbalance management techniques to develop a powerful model to identify fraudulent transactions, the dataset must be balanced. In order to address the issue of class imbalance, this study also seeks to give a comparative analysis of several individual and hybrid resampling techniques. In several experimental studies, the effectiveness of various resampling techniques in combination with classification approaches has been compared. In this study, it is found that the hybrid resampling methods perform well for machine learning classification models compared to deep learning classification models. The experimental results show that the best accuracy for the Random Forest (RF); Logistic Regression; K-Nearest Neighbors (KNN); Decision Tree (DT), and Gaussian Naive Bayes (NB) classifiers are 99,99%; 94,61%; 99.96%; 99,98%, and 91,47%, respectively. The comparative results show that the RF outperforms with high performance parameters (accuracy, recall, precision and f score) better than NB; RF; DT and KNN. RF achieve the minimum loss values with all resampling techniques, and the results, when utilizing the proposed models on the entire skewed dataset, achieved preferable outcomes to the unbalanced dataset. Furthermore, the PyTorch framework achieves higher prediction accuracy for the federated learning model than the TensorFlow federated framework but with more computational time.

Comparison of CNN-based methods for yoga pose classification

Yoga is an exercise developed in ancient India. People perform yoga in order to have mental, physical, and spiritual benefits. While yoga helps build strength in the mind and body, incorrect postures might result in serious injuries. Therefore, yoga exercisers need either an expert or a platform to receive feedback on their performance. Since access to experts is not an option for everyone, a system to provide feedback on the yoga poses is required. To this end, commercial products such as smart yoga mats and smart pants are produced; Kinect cameras, sensors, and wearable devices are used. However, these solutions are either uncomfortable to wear or not affordable for everyone. Nonetheless, a system that employs computer vision techniques is a requirement. In this paper, we propose a deep-learning model for yoga pose classification, which is the first step of a quality assessment and personalized feedback system. We introduce a wavelet-based model that first takes wavelet transform of input images. The acquired subbands, i.e., approximation, horizontal, vertical, and diagonal coefficients of the wavelet transform are then fed into separate convolutional neural networks (CNN). The obtained probability results for each group are fused to predict the final yoga class. A publicly available dataset with 5 yoga poses is used. Since the number of images in the dataset is not enough for a deep learning model, we also perform data augmentation to increase the number of images. We compare our results to a CNN model and the three models that employ the subbands separately. Results obtained using the proposed model outperforms the accuracy output achieved with the compared models. While the regular CNN model has 61% and 50% accuracy for the training and test data, the proposed model achieves 91% and 80%, respectively.