Browsing by Author "Karpagavalli S"

Now showing 1 - 20 of 43

A ROBUST DEEP LEARNING MODEL FOR SKIN DISEASE CLASSIFICATION
(2023) Kalaivani A; Karpagavalli S
newline
ADVANCED DOMAIN ADAPTATION FOR SKIN DISEASE SEGMENTATION AND CLASSIFICATION USING BOOTSTRAPPING OF FINE-TUNED DEEP LEARNER (Article)
(Springer, 2024-04) Kalaivani A; Karpagavalli S
In medical diagnostic systems, the most challenging task is to segment and classify the varieties of skin disorders from dermoscopic images. For this purpose, Bootstrapping of Fine-tuned Segmentation and Classification Network (BF-SegClassNet) model was designed, which uses (i) cycle-Generative Adversarial Network (GAN) as domain adaptation, (ii) modified SegNet as segmentation and (iii) fine-tuned ResNet18 with Bootstrapping as classification. But, the efficiency of cycle-GAN was degraded if the source domain differs largely from the target domain. Hence, in this article, a Fuzzy Transfer Learning (FTL) model is developed based on fuzzy logic as domain adaptation. In this model, 2 different stages are performed such as training and adaptation. During the training stage, the source labeled data is used to build the Fuzzy Inference System (FIS), which extracts information from the source and transfers it to the target domain. The fuzzy sets and fuzzy rules created by an Adhoc Data-Driven Learning (ADDL) activity are included in the FIS. The created source FIS and the target data are used in the adaptation stage to adapt the fuzzy rule and the fuzzy rule base from the FIS to extract dissimilarities in the data and help bridge the contextual gap between the source and target. Thus, this FTL model is applied instead of cycleGAN to create more samples, which are further partitioned and classified by the BF-SegClassNet model efficiently. Finally, the testing outcomes exhibit that the FTL model attains a mean accuracy of 98.08% for the HAM dataset compared to the other GAN models.
ANALYSIS OF TAMIL CHARACTER WRITINGS AND IDENTIFICATION OF WRITER USING SUPPORT VECTOR MACHINE
(IEEE International Conference on Advanced Communication Control and Computing Technologies (ICACCCT), 2014 indexed in IEEE Xplore Digital Library, 2014) Thendral T; Vijaya M S; Karpagavalli S
Distinctive Handwriting is a thought provoking task in writer identification. The style and shape of the letters written by the same writer may vary and entirely different for different writers. Alphabets in the handwritten text may have loops, crossings, junctions, different directions and so on. Therefore exact prediction of individual based on his/her handwriting is highly complex and challenging task. This paper proposes a new model for learning the writer's identity constructed on Tamil handwriting. Handwritten documents written by the writers are scanned and segmented into words. Words are further segmented into characters for character level writer identification. The character writings in Tamil are analyzed and their describing features are defined. The Writer identification problem is formulated as classification task and a pattern classification technique namely Support Vector Machine has been employed to construct the model. It has been reported about 90. 6% of prediction accuracy by RBF kernel based classification model in character level writer identification.
AUTOMATIC SPEECH RECOGNITION: ARCHITECTURE, METHODOLOGIES, CHALLENGES - A REVIEW
(International Journal of Advanced Research in Computer Science, 2011-11) Karpagavalli S; Deepika R; Kokila P; Usha Rani K; Chandra E
For more than three decades, a great amount of research was carried out on various aspects of speech signal processing and its applications. Highly successful application of speech processing is Automatic Speech Recognition (ASR). Early attempts to ASR consisted of making deterministic models of whole words in a small vocabulary and recognizing a given speech utterance as the word whose model comes closest to it. The introduction of Hidden Morkov Models (HMMs) in the early 1980 provided much more powerful tool for speech recognition. And the recognition can be done for continuous speech using large vocabulary, in a speaker independent manner. Today many products have been developed that successfully utilize ASR for communication between human and machines. Performance of speech recognition applications deteriorates in the presence of reverberation and even low levels of ambient noise. Robustness to noise, reverberation and characteristics of the transducer is still an unsolved problem that makes the research in the area of speech recognition still very active. A detailed study on ASR carried out and presented in this paper that covers the basic model of speech recognition, applications
AUTOMATIC TAG RECOMMENDATION FOR JOURNAL ABSTRACTS USING STATISTICAL TOPIC MODELING
(Springer Advances in Intelligent Systems and Computing(AISC Series), 2015) Anupriya P; Karpagavalli S
Topic modeling is a powerful technique for unsupervised analysis of large document collections. Topic models conceive latent topics in text using hidden random variables, and discover that structure with posterior inference. Topic models have a wide range of applications like tag recommendation, text categorization, keyword extraction and similarity search in the broad fields of text mining, information retrieval, statistical language modeling. In this work, a dataset with 200 abstracts fall under four topics are collected from two different domain journals for tagging journal abstracts. The document model is built using LDA (Latent Dirichlet Allocation) with Collapsed Variational Bayes (CVB0) and Gibbs sampling. Then the built model is used to find appropriate tag for a given abstract. An interface is designed to extract and recommend the tag for a given abstract.
BOOTSTRAPPING OF FINE-TUNED SEGMENTATION AND CLASSIFICATION NETWORK FOR EPIDERMIS DISORDER CATEGORIZATION
(Springer Link, 2023-07-25) Kalaivani A; Karpagavalli S
As all people have been affected by various skin related illnesses, categorization of skin disorders has become prominent in recent healthcare system. To identify and categorize skin related syndromes, many transfer learning frameworks were used. Amongst, a Fine-tuned Segmentation and Classification Network (F-SegClassNet) achieved better efficacy by using the novel unified loss function. Nonetheless, it was not apt for the datasets that lack in training images. Hence in this article, Bootstrapping of F-SegClassNet (BF-SegClassNet) model is proposed which solves the imbalanced images in the training set via generating the group of pseudo balanced training batches relying on the properties of the considered skin image dataset. This model fits the distinct abilities of Deep Convolutional Neural Network (DCNN) classifier so that it is highly useful for classifying the skin disorder image dataset with a highly imbalanced image data distribution. According to the Bootstrpping, better tradeoff between simple and complex image samples is realized to make a network model that is suitable for automatic skin disorders classification. In this model, statistics across the complete training set is calculated and a new subset is produced that retains the most essential image samples. So, the skin images are segmented and categorized by this new model to identify the varieties of epidermis infections. At last, the testing outcomes exhibits BF-SegClassNet-model accomplishes the mean accuracy with 96.14% for HAM dataset which is compared to state-of-the-art models.
BREAST CANCER CLASSIFICATION USING SUPPORT VECTOR MACHINE AND GENETIC PROGRAMMING
(International Journal of Innovative Research in Computer and Communication Engineering (IJIRCCE), 2013-09) Menaka K; Karpagavalli S
Breast cancer is one of the most leading causes of death among women. The early detection of abnormalities in breast enables the radiologist in diagnosing the breast cancer easily. Efficient tools in diagnosing the cancerous breast will help the medical experts in accurate diagnosis and timely treatment to the patients. In this work, experiments carried out using Wisconsin Diagnosis Breast Cancer database to classify the breast cancer either benign or malignant. Supervised learning algorithm Support Vector Machine (SVM) with kernels like Linear, Polynomial and Radial Basis Function and evolutionary algorithm Genetic Programming are used to train the models. The performance of the models are analysed where genetic programming approach provides more accuracy compared to Support Vector Machine in the classification of breast cancer and seems to be an fast and efficient method.
CLASSIFICATION OF LUNG DISEASE USING LOCAL AND GLOBAL DESCRIPTORS
(International Journal of Computer Applications, 2016-02) Pradeebha R; Karpagavalli S
Recent trends indicate that instances of chronic respiratory diseases are on the rise in India mainly due to vehicular pollution, air and dust pollution, habit of smoking and also increased population. A World Health Organization report indicates that India has a ranking number one in the world for lung disease deaths. Respiratory diseases like asthma, chronic obstructive pulmonary disease (COPD), Interstitial Lung Disease (ILD), pneumonia, tuberculosis (TB) are emerging as most important health problems in the country. The proposed work is aimed at establishing more advanced diagnostic strategy for lung diseases using CT scan images. Lung diseases such as Emphysema, Pneumonia, Bronchitis are classified using CT scan images which is collected from National Biomedical Imaging Archive (NBIA). A total of 366 images are used, out of which 300 images are used for training and 66 images are used for testing. The classification task carried out with classifier support vector machine (SVM) using Histogram of Oriented Gradient (HOG) –global descriptors and Local Binary Pattern (LBP) – local descriptors. The performance of the model built using Support Vector Machine indicates that it is effective in the prediction of lung disease with 98% predictive accuracy.
CLASSIFICATION OF SEED COTTON YIELD BASED ON THE GROWTH STAGES OF COTTON CROP USING MACHINE LEARNING TECHNIQUES
(IEEE Xplore and IEEE CS Digital Library, 2010-06) Jamuna K S; Karpagavalli S; Vijaya M S; Revathi P; Gokilavani S; Madhiya E
Cotton, popularly known as "White Gold" has been an important commercial crop of national significance due to the immense influence of its rural economy. Cotton seed is an important and critical link in the chain of agricultural activities extending farmer industry linkage. Cotton yield is associated with high quality seed as the seed contains in itself the blue print for the agrarian prosperity in incipient form. Transfer of technology to identify the quality of seeds is gaining importance. Hence this work employs machine learning approach to classify the quality of seeds based on the different growth stages of the cotton crop. Machine learning techniques - Naïve Bayes Classifier, Decision Tree Classifier and Multilayer Perceptron were applied for training the model. Features are extracted from a set of 900 records of different categories to facilitate training and implementation. The performance of the model was evaluated using 10 -fold cross validation. The results obtained show that Decision Tree Classifier and Multilayer Perceptron provides the same accuracy in classifying the seed cotton yield. The time taken to build the model is higher in Multilayer Perceptron as compared to the Decision Tree Classifier.
A DEEP ENSEMBLE MODEL FOR AUTOMATED MULTICLASS CLASSIFICATION USING DERMOSCOPY IMAGES
(IEEE, 2022-03-31) Kalaivani A; Karpagavalli S
In medical diagnosis, manual skin tumor treatment is time consuming and exclusive, it is important to create computerized analytic strategies that can accurately classify skin lesions of many stages. A completely automatic way to classify skin lesions of many categories has been presented. Automatic dissection of skin lesions and isolation are two major and related functions in the diagnosis of computer-assisted skin cancer. Even with their widespread use, deep learning models are typically only intended to execute a single task, neglecting the potential benefits of executing both functions simultaneously. The Bootstrapping Ensembles based Convolutional Neural Networks (BE-CNN) model is proposed in this paper for the separation of skin lesions simultaneously and for classification. A Compute-Intensive Segmentation Network (CI-SN), comprise this model (improved-SN). On one hand, Compute-Intensive Segmentation Network creates uneven lesion covers that serves as a pre-bootstrapping, allowing it to reliably find and classify skin lesions. Both division and arrangement networks, in this approach, mutually transmit assistance and experience each other in a bootstrapping manner. However, to deal with the challenges posed by class inequality and simple pixel inequality, a novel method in segmentation networks is proposed. On the ISIC-HAM 10000 datasets, the proposed BE-CNN model is evaluated and found that it achieves mean skin lesion classification accuracy of 93.8 percentile, which is higher than the function of the separation of skin lesions representing the modern condition and stages techniques. Proposed outcomes demonstrate that via preparing a bound together model to execute the two tasks in a non-stop bootstrapping strategy, it is feasible to work on the presentation of skin sore division and grouping simultaneously.
DEEP NEURAL NETWORK OPTIMIZATION FOR SKIN DISEASE CLASSIFICATION FORECAST ANALYSIS
(IEEE, 2022-03-26) Kalaivani A; Karpagavalli S
Skin lesions are a prevalent condition that causes misery, many of which can be severe, for millions of individuals worldwide. Consequently, Deep learning seems to be an increasingly popular approach in recent years, and it may be a strong tool in difficult, earlier domains, specifically in health science, which is now dealing with a number of medical resources. In this paper, presented an interactive dermoscopy images diagnosis framework based on an gathering of intelligent deep learning model system for image classification to make advances their person accuracies within the prepare of classifying dermoscopy pictures into several classes such as melanoma, keratosis and nevus when we have not sufficient annotated images to train them on. We integrate the classification layer results for two distinct deep neural network designs to obtain excellent classification accuracy. More precisely, we combining robust convolutional neural networks (CNNs) into a unified structure, with the final classification relying on the weighted outcome of the respective CNNs by predictive ensemble methods and fine-tuning classifiers utilizing ISIC2019 images. Furthermore, the outliers and the substantial class imbalance are handled in order to improve the categorization of the disease. The experimental reveal that the framework produced result that are comparable to other models of conventional art. A substantial improvement in accuracy of 96.2 percentage indicated the efficiency of the proposed Predictive Ensemble Deep Convolutional Neural Networks Classifier (PE-DCNN Classifier) model and this study effectively built a system with all the important features.
DIABETIC RETINAL EXUDATES DETECTION USING EXTREME LEARNING MACHINE
(CSI Annual Convention and International Conference on Emerging ICT for Bridging the Future and published in Springer Advances in Intelligent Systems and Computing(AISC Series), 2015) Asha P R; Karpagavalli S
Diabetic Retinopathy is a disorder of the retina as a result of the impact of diabetes on the retinal blood vessels. It is the major cause of blindness in people like age groups between 20 & 60. Since polygenic disorder proceed, the eyesight of a patient may commence to deteriorate and causes blindness. In this proposed work, the existence or lack of retinal exudates are identified using Extreme Learning Machine(ELM). To discover the occurrence of exudates features like Mean, Standard deviation, Centroid and Edge Strength are taken out from Luv color space after segmenting the Retinal image. A total of 100 images were used, out of which 80 images were used for training and 20 images were used for testing. The classification task carried out with classifier extreme learning machine (ELM). An experimental result shows that the model built using Extreme Learning Machine outperforms other two models and effectively detects the presence of exudates in retina
DIABETIC RETINAL EXUDATES DETECTION USING MACHINE LEARNING TECHNIQUES
(International Conference on Advanced Computing & Communication Systems, held at Sri Eshwar College of Engineering, Coimbatore during 5-7 January 2015 and published in the conference proceedings, indexed in IEEE Xplore Digital Library., 2015-01-05) Asha P R; Karpagavalli S
Diabetic Retinopathy (DR) is an eye filled illness caused by the complication of polygenic disease and that is to be detected accurately for timely treatment. As polygenic disease progresses, the vision of a patient could begin to deteriorate and leads to blindness. In this proposed work, the presence or absence of retinal exudates are detected using machine learning (ML) techniques. To detect the presence of exudates features like Mean, Standard deviation, Centroid and Edge Strength are extracted from Luv color space after segmenting the Retinal image. A total of 100 images were used, out of which 80 images were used for training and 20 images were used for testing. The classification task carried out with classifiers like Naive bayes (NB), Multilayer Perceptron (MLP) and Extreme Learning Machine (ELM). Experimental results shows that the model built using Extreme Learning Machine outperforms other two models and effectively detects the presence of exudates in retinal images.
DISCOVERING TAMIL WRITER IDENTITY USING GLOBAL AND LOCAL FEATURES OF OFFLINE HANDWRITTEN TEXT
(International Review on Computers and Software (IRECOS), 2013) Thendral T; Vijaya M S; Karpagavalli S
Writer identification is the process of identifying the individual based on their handwriting. Handwriting exhibits behavioral characteristics of an individual and has been considered as unique. The style and shape of the letters written vary slightly for same writer and entirely different for different writers. Also alphabets in the handwritten text may have loops, crossings, junctions, different directions etc. Hence accurate prediction of individual based on his/her handwriting is highly complex and challenging task. This paper proposes a new model for discovering the writer’s identity based on Tamil handwriting. Writer identification problem is formulated as classification task and a pattern classification technique namely Support Vector Machine has been employed to construct the model. It has been reported about 93.8% of prediction accuracy by RBF kernel based classification model.
ELECTROCARDIOGRAM BEAT CLASSIFICATION USING SUPPORT VECTOR MACHINE AND EXTREME LEARNING MACHINE
(Springer, 2014) Banu Priya C V; Karpagavalli S
The Electrocardiogram (ECG) is of significant importance in assessing patients with abnormal activity in their heart. ECG Recordings of the patient taken for analyzing the abnormality and classify what type of disorder present in the heart functionality. There are several classes of heart disorders including Premature Ventricular Contraction (PVC), Atrial Premature beat (APB), Left Bundle Branch Block (LBBB), Right Bundle Branch Block (RBBB), Paced Beat (PB), and Atrial Escape Beat (AEB).To analyze ECG various feature extraction methods and classification algorithms are used. The proposed work employed discrete wavelet transform (DWT) in feature extraction on ECG signals obtained from MIT-BIH Arrhythmia Database. The Machine Learning Techniques, Support Vector Machine (SVM) and Extreme Learning Machine (ELM) have been used to classify four types of heart beats that include PVC, LBBB, RBBB and Normal. The Performance of the classifiers are analyzed and observed that ELM-Radial Basis Function Kernel taken less time to build model and out performs SVM in predictive accuracy.
EMAIL SPAM FILTERING USING SUPERVISED MACHINE LEARNING TECHNIQUES
(International Journal of Advanced Research in Computer Science, 2011-12) Christina V; Karpagavalli S; Suganya G
E-mail spam, known as unsolicited bulk Email (UBE), junk mail, or unsolicited commercial email (UCE), is the practice of sending unwanted e-mail messages, frequently with commercial content, in large quantities to an indiscriminate set of recipients. Spam is prevalent on the Internet because the transaction cost of electronic communications is radically less than any alternate form of communication. There are many spam filters using different approaches to identify the incoming message as spam, ranging from white list / black list, Bayesian analysis, keyword matching, mail header analysis, postage, legislation, and content scanning etc. Even though we are still flooded with spam emails everyday. This is not because the filters are not powerful enough, it is due to the swift adoption of new techniques by the spammers and the inflexibility of spam filters to adapt the changes. In our work, we employed supervised machine learning techniques to filter the email spam messages. Widely used supervised machine learning techniques namely C 4.5 Decision tree classifier, Multilayer Perceptron, Naïve Bayes Classifier are used for learning the features of spam emails and the model is built by training with known spam emails and legitimate emails. The results of the models are discussed.
EXPERT AUTOMATED SYSTEM FOR PREDICTION OF MULTI-TYPE DERMATOLOGY SICKNESSES USING DEEP NEURAL NETWORK FEATURE EXTRACTION APPROACH
(IJISAE, 2023) Kalaivani A; Karpagavalli S; Kamal, Gulati
One of the most prevalent illnesses on the planet is skin issues. Due to the complexity of types of skin, and hair types, it is difficult to evaluate it despite its popularity.Consequently, skin conditions pose a serious public health danger. When they reach the invasive stage of evolution, they become harmful. Medical professionals are very concerned about dermatological disorders. The number of people who suffer from skin illnesses is growing substantially as a result of rising pollution and bad food. People frequently ignore the early indications of skin conditions. A hybrid approach can minimize human judgment, producing positive results quickly. A thorough examination suggests that frameworks for recognizingvarious skin disorders may be built using deep learning techniques. To find skin illnesses, it is necessary to distinguish between theskin and non-skin tissue. Through the use of feature extraction-baseddeep neural network approaches, a classification system for skin diseases was established in this study. The main goal of this system is to anticipate skin diseases accurately while also storing all relevant state data efficiently and effectively for precise forecasts. The significant issues have been addressed, and a unique, feature extraction-based deep learning modelis introduced to assist medical professionals in properly detecting the type of skin condition.The pre-processing stage is when the inputdataset is first supplied, helping to clear the image of any undesired elements. Then, for the training phase, the proposed Feature Extraction Based Deep Neural Network (FEB-DNN) is fed the features collected from each of the pre-processed frames. With the use of measured parameters, the classification system categorizesincoming treatment data as various skin conditions. Finding the ideal weight values to minimizetraining error is crucial while learning the proposed framework. In this study, an optimization strategy is used to optimizethe weight in the structure. Based on the feature extraction approach, the suggested multi-type framework for diagnosing skin diseases has a 91.88% of accuracyrate for the HAM image dataset and identifies several skin disorder subtypes than the earlier models thatcan aid in treatment response and decision-makingwhich alsohelp doctors make an informed decision.
EXPLORATION OF DEEP GENERATIVE MODELS FOR ADDRESSING DATA IMBALANCE IN COMPUTER VISION (Conference Paper)
(Springer Science and Business Media Deutschland GmbH, 2025-02-25) Priyadharshini A; Karpagavalli S
Computer vision tasks primarily depend on image data collection, preprocessing of the images, and the feature extraction steps to accomplish the task. When the collected images count is highly imbalanced within a category or between multiple categories involved, the anticipated tasks cannot be achieved. Data imbalance is a communal tricky problem that occurs in acquiring image data collection in specific complex real-world scenarios such as medical diagnosis, fraud detection, manufacturing defect detection, object detection, biological research etc., are inevitable. Several traditional methods for class imbalance are available but more advanced data generation models show promising results. The effectiveness of the computer vision algorithms has an inverse relationship to the quality and quantity of the dataset. In recent years, data generation models like Generative Adversarial Network, Variational Autoencoders, have gained enormous attention from researchers across domains, due to their potential for learning to generate images and their effectiveness in restoring balance in imbalanced datasets. They not only enhance the dataset volume but also contribute to the generalization and robustness of the machine learning algorithms. Moreover, data generation models play a vital role in improving classification accuracy, object detection and semantic segmentation tasks in deep learning. The survey is done with an intention to deliver a comprehensive study of data generation models, their learning representations from data, and their efficiency to address various challenges related to dataset quality, size and diversity.
A HIERARCHICAL APPROACH IN TAMIL PHONEME CLASSIFICATION USING SUPPORT VECTOR MACHINE
(Indian Journal of Science and Technology, 2015-12) Karpagavalli S; Chandra E
Most of the speech recognition systems are designed based on the sub-word unit phoneme which is the basic sound unit of a language. In the proposed work, a novel hierarchical approach based phoneme classification task has been carried out to reduce time complexity and search space. Hierarchical classification of set of Tamil phonemes has been done in three levels. Phoneme boundaries of the given speech utterance are identified using Spectral Transition Measure (STM) and phonemes are separated. Mel-Frequency Cepstral Coefficients (MFCC) are extracted for each phoneme represented by 9 frames including the contextual frames of corresponding phoneme. In each hierarchical level, different number of models is built using Support Vector Machine (SVM) for classifying each phoneme group/phoneme. It is observed from the results that in hierarchical approach phoneme group recognition rate at level 1 and 2 has greatly improved compared to flat classification model. Complexity of search space is significantly reduced at level 2 and level 3 contrasts to flat phoneme classification model. Hierarchical phoneme classifier can be very well employed in phoneme recognition task which is useful in applications such as spoken term detection, out-ofvocabulary detection, named entity recognition, spoken document retrieval.
AN INTERACTIVE TOOL FOR YARN STRENGTH PREDICTION USING SUPPORT VECTOR REGRESSION
(IEEE Xplore, 2010-05-06) Selvanayaki M; Vijaya M S; Jamuna K S; Karpagavalli S
Cotton, popularly known as White Gold has been an important commercial crop of National significance due to the immense influence of its rural economy. Transfer of technology to identify the quality of fibre is gaining importance. The physical characteristics of cotton such as fiber length, length distribution, trash value, color grade, strength, shape, tenacity, density, moisture absorption, dimensional stability, resistance, thermal reaction, count, etc., contributes to determine the quality of cotton and in turn yarn strength. In this paper yarn strength prediction has been modeled using regression. Support Vector regression, the supervised machine learning technique has been employed for predicting the yarn strength. The trained model was evaluated based on mean squared error and correlation coefficient and was found that the prediction accuracy of SVR based model, the intelligence reasoning method is higher compared with the traditional statistical regression, the least square regression model.