Suganthi D
@Saveetha Institute of Medical and Technical Sciences(SIMATS), Saveetha Nagar, Thandalam, Chennai, India
Assistant Professor(SG) and Computer Science
Saveetha College of Liberal Arts and Sciences (SCLAS),
Very passionate, dedicated, and hard-working person.
EDUCATION
B.Sc(Mathematics).,
MCA.,
M.Phil.,
M.E.,
Science)
Scopus Publications
Scholar Citations
Scholar h-index
Scholar i10-index
Scopus Publications
Varuna Kumara, Durgesh M Sharma, J. Samson Isaac, S. Saravanan, D. Suganthi, and Sampath Boopathi
IGI Global
The book explores the use of artificial intelligence (AI) in power monitoring systems for SMEs to enhance energy efficiency, reduce operational costs, and ensure sustainability. It discusses current energy challenges faced by SMEs, emphasizes real-time monitoring, and highlights the benefits of AI integration. The chapter details the components of an AI-integrated power monitoring system, including data acquisition, analysis, and control strategies. It examines AI techniques like machine learning, deep learning, and predictive analytics for identifying energy usage patterns. The chapter also discusses successful cases of SMEs using AI-based systems, highlighting their optimization of energy consumption and reduced costs.
Dileep P, M. Durairaj, Sharmila Subudhi, V V R Maheswara Rao, J. Jayanthi, and D Suganthi
Springer Science and Business Media LLC
D. Saravanan, R. Parthiban, G. Arunkumar, D. Suganthi, R. Revathi, and U. Palani
Wiley
Chamandeep Kaur, Mohammed Saleh Al Ansari, Vijay Kumar Dwivedi, and D. Suganthi
Wiley
L. Rajeshkumar, J. Rachel Priya, Konatham Sumalatha, G. Arunkumar, D. Suganthi, and D. Saravanan
Wiley
Chamandeep Kaur, Mohammed Saleh Al Ansari, Vijay Kumar Dwivedi, and D. Suganthi
Wiley
Vishwa Priya V, Padmini Chattu, K. Sivasankari, Dnyaneshwar Tukaram Pisal, Bantupalli Renuka Sai, and D. Suganthi
IEEE
Modern healthcare relies on medical imaging to diagnose and cure diseases. Convolution Neural Networks (CNNs) have excelled at image categorization, and their use in medical imaging could change illness diagnosis. This study examines CNN's medical picture categorization performance. This study used X-rays, CT scans, MRIs, and histopathology slides. Traditional CNNs, transfer learning models, and bespoke networks were constructed and trained. The training data included lung ailments, cancer detection, bone fractures, and neurological disorders. Our investigations showed that CNNs can extract complex characteristics from medical photos, improving classification accuracy. Transfer learning, where pre-trained models were fine-tuned on medical data, performed well with 94.8% accuracy. We used cutting-edge data augmentation and attention strategies to improve model generalization and interpretability. In addition to high classification accuracy, we studied model explainability using gradient-based methods and visualization to highlight medical picture regions of relevance that influenced model predictions. Building trust with medical practitioners and understanding deep learning model decision-making requires interpretability. In clinical settings, CNN deployment raises ethical and practical issues such as data privacy, model robustness, and regulatory compliance. According to our research, CNNs may increase medical picture classification accuracy and speed. X-rays had the best accuracy at 95.2%, followed by CT scans at 92.7% and MRIs at 94.1%. Histopathological slides exhibited 88.6% accuracy; however, this shows that CNNs can diagnose diseases from microscopic images.
Jose Reena K, Charul Nigam, G. Kirubasri, S. Jayachitra, Anurag Aeron, and D. Suganthi
IEEE
In an era when edge computing rapidly evolves, this study addresses the significant difficulty of real-time object identification on resource-constrained edge devices. We provide a unique neural network model for edge scenarios in this paper. Our model covers object recognition system accuracy, computational efficiency, and speed. Object detection methods that need processing are impractical for edge computing. This paper proposes a Mobile Neural Network tailored to the limitations. Pruning, which decreases model size by 30%, and quantization make the model efficient. On several edge devices, the model had an average accuracy of 92% on Dataset 1 and 89% on Dataset 2. The model's 40-millisecond inference time and 2.3-watt power consumption were impressive. It outperformed standard CNN models and edge-optimized algorithms like YOLOv3 and SSD MobileNet under identical conditions. This study shows that the proposed paradigm can revolutionise edge computing real-time object detection. The model is suitable for high-responsiveness, energy-efficient applications.
Cuddapah Anitha, Shivali Devi, Vinay Kumar Nassa, Mahaveerakannan R, Kingshuk Das Baksi, and Suganthi D
Anapub Publications
Data from environmental monitoring can be used to identify possible risks or adjustments to ecological patterns. Early detection reduces risks and lessens the effects on the environment and public health by allowing for prompt responses to ecological imbalances, pollution incidents, and natural disasters. Decision-making and analysis can be done in real time when Artificial Intelligence (AI) is integrated with Unmanned Aerial Vehicles (UAV) technology. With the help of these technologies, environmental monitoring is made possible with a more complete and effective set of tools for assessment, analysis, and reaction to changing environmental conditions. Multiple studies have shown that forest fires in India have been happening more often recently. Lightning, extremely hot weather, and dry conditions are the three main elements that might spontaneously ignite a forest fire. Both natural and man-made ecosystems are affected by forest fires. Forest fire photos are pre-processed using the Sobel and Canny filter. A Convolutional Neural Network (CNN)–based Forest Fire Image Classification Network (DFNet) using the publicly accessible Kaggle dataset is proposed in this study. The suggested DFNet classifier's hyperparameters are fine-tuned with the help of Spotted Hyena Optimizer (SHO). With a performance level of 99.4 percent, the suggested DFNet model outperformed the state-of-the-art models, providing substantial backing for environmental monitoring.
Balasubramani Ramesh, Sudhagar Dhandapani, Sanda Sri Harsha, Naheem Mohammed Rahim, Nanda Ashwin, Duraisamy Suganthi, and Rengaraj Gurumoorthy Vidhya
Akademia Baru Publishing
Brain tumour identification and categorization are critical for diagnosis and treatment. This work uses preprocessing and classification algorithms to discover and categories brain tumours. Gaussian smoothing reduces noise and improves image quality, Genetic Algorithms select and optimize features, Deep Learning-Based Segmentation accurately segments tumours, Local Binary Patterns (LBP) extract texture features, and K-Nearest Neighbors classify tumours. Gaussian smoothing reduces noise and improves brain imaging data. Genetic Algorithms extract the most relevant and discriminative features from preprocessed photos. A Deep Learning-Based Segmentation model accurately segments brain tumour regions using these features. After segmentation, Local Binary Patterns (LBP) extract tumour texture features. The K-Nearest Neighbors (KNN) method classifies tumours using these texture features to capture tumour spatial patterns. Our suggested brain tumour detection and classification method combines several techniques to increase accuracy and reliability. Gaussian smoothing and LBP feature extraction improve feature discrimination. Deep Learning-Based Segmentation and the KNN classifier ensure precise tumour location and robust classification. The proposed method will be tested on brain scans including tumour areas. Classification performance measures include accuracy, sensitivity, specificity, and AUC. This work will improve brain tumour detection and classification methods for more accurate diagnoses and treatment planning. The primary goal of the work is to enhance brain tumour identification and categorization using pre processing, classification algorithms, and advanced techniques, ultimately improving diagnosis and treatment outcomes.
D. Sengeni, G. Padmapriya, S. Sagar Imambi, D. Suganthi, Ashish Suri, and S. Boopathi
IGI Global
The handling of genetic residues, biohazardous waste, and other non-shrinkable materials such as e-waste is essential for the preservation of Mother Nature and human wellbeing. This chapter will bring forth different methods of managing biomedical wastes, inspecting the roots of the spread of the hazardous virus, and protecting the front-line warriors. To properly react to this disaster, each administration must create a backup strategy based on regional circumstances and the seriousness of the coronavirus's spread. Technology is being used to reduce the need for in-person medical appointments and save on the cost of treatment, which is a boon for human civilization in the context of the pandemic. The Central Pollution Control Board's existing rules provide a framework for managing healthcare waste with respect for the safety of the public. Bio-waste management strategies using IoT and IoMT techniques are illustrated in this chapter.
Sameer Yadav, P Nithya, T. Manikumar, D. Suganthi, P. Vijayaragavan, and Savanam Chandra Sekhar
IEEE
Information and communication technology (ICT) enables the smart city to share data with the public and provide superior services to its residents. Smart technology has the ability to greatly improve the efficiency and quality of rubbish collection on a global scale, which is why waste collection is an integral part of smart city services. Garbage is taken out even though it's not full, city services are overused, and petrol is wasted. Garbage collection in smart cities has two major flaws: high prices and low efficiency. Recycling is the answer since it lessens the amount of trash that needs to be disposed of and saves valuable space in landfills. Estimates suggest that humans can produce more rubbish than ever before, despite the fact that recycling rates are on the rise. Creating algorithms that can learn from data and apply that data to new situations is a major goal in machine learning. Preprocessing, feature selection, and model training are the three components of the proposed method. The proposed method employs normalization for preprocessing. It uses a clustering-based approach to extract and select features. The LSTM-AE technique is used to assess the model's efficiency. When compared to conventional approaches like LSTM and GRU, the proposed method performs exceptionally well.
Ranjeet Kumar, Santosh Kumar Behera, Kalpana K, Amit V Kadam, V. Ashok, and D. Suganthi
IEEE
Recent increases in healthcare spending have made early disease detection a pressing concern. The growth of populations has a direct correlation with the rising death toll from breast cancer. Breast cancer ranks second among the deadliest cancers the proposed approach knows about. An autonomous disease detection system minimizes the likelihood of fatalities by providing a rapid response from medical personnel who can effectively treat the condition. The proposed approach entails three steps: preprocessing, segmentation, and model training. The proposed approach employed a combination of Contrast Limited Histogram Equalization and Fuzzy Histogram Equalization as a preprocessing technique. More and more comprehensive segmentation regions are produced when numerous segmentation regions are connected. After accumulating the attributes, the models are trained with RA-UNet. The proposed technique outperforms its counterparts, including RLM and U-Net. The probability of this strategy succeeding is 97.36 percent.
Sameer Yadav, K Santosh Reddy, Bhaskar Marapelli, Bhola Khan, D. Suganthi, and Shiv Ashish Dhondiyal
IEEE
The impact of the information technology sector is being seen in every facet of today's globalized economy. In order to progress, countries like India require greater assistance in the agricultural sector. Farmers and the government can both benefit from accurate price forecasts. Vegetable price prediction is difficult; thus, the proposed system used the neural network features of self-adaptation, self-study, and high fault tolerance to construct a Backpropagation neural network model. The neural network was used to establish a model for making forecasts. Preprocessing, feature selection, and evaluating the model's performance are the three stages that make up the suggested method. Preprocessing typically involves normalization. PCA is used to pick features. It employs GC-A-LSTM for measuring the model's efficacy. The proposed method is contrasted with LSTM and A-LSTM, two established approaches. When compared to more simplistic approaches, it excels.
S L Prathapareddy, Neelam Sharma, Jayalakshmi V, D. Suganthi, S.Sathya Naveena, and Rajanish Kumar Kaushal
IEEE
Increased interest in video analysis technology, and in particular automatic human activity recognition, has arisen in response to the rising requirements of a wide range of applications, including surveillance, entertainment, and healthcare systems. Using automated reporting, authorities might be alerted to the presence of a possible criminal or dangerous actor, such as a person hanging about an airport or railway station with a bag. Similarly, activity identification can improve human-computer interaction in the gaming industry by, say, automatically identifying the motions of different players in a tennis match so that an avatar can assume control of the game on the player's behalf. Steps one through four of the proposed procedure are preprocessing the data, segmenting the data, selecting features, and training the model. Preprocessing videos is used to observe people's actions. Background subtraction and GMM both play a role in the segmentation process. Parameters such as speed, measurement, and movement play a significant role in feature extraction. The models are then trained with LSTM-RNN after feature extraction. The proposed method outperforms the two most used alternatives, LSTM and RNN.
Mathur Nadarajan Kathiravan, Shivangi Bande, Sadhana Tiwari, Ashok Kumar Koshariya, Manju M, and D Suganthi
IEEE
Changes in land use over the past few decades have left their mark on Earth's surface. Key ecosystem services provided by soils are currently being altered due to global environmental changes. In light of these factors, monitoring equipment is required to maintain a healthy ecological status and advance soil conservation on a local, regional, and global scale. This system demonstrates the potential utility of remote sensing in collecting soil properties. Methodologies based on various remote sensing devices and classification strategies have been developed for assessing soil characteristics. Soil analysis by remote sensing is a relatively new area of study in the agricultural sciences. The majority of people in India rely on agriculture as a means of subsistence and economic survival. The low cost of yield prediction has led to a decline in agricultural output in recent years. Predicting yields is challenging because of the many variables at play, including as soil, rainfall, and fertilizers. The quality of the soil is crucial for agricultural output. MSC and SNV are used in the suggested method for preprocessing. LDA is used for feature selection. The proposed method employs FGCRNN to train the envisioned model. The proposed model is evaluated in comparison to the CNN and GRU models. Its performance is superior to that of the other two types.
Mahesh Prasanna K, Mary Selvan, Deepa R, D. Suganthi, B. Uma Maheswari, and Parul Madan
IEEE
In this work, a proof of concept is presented, which shows that EIT images reconstructed by EIDORS using methods based on the Finite Element Method can be transformed into less smooth versions using a combination of unsupervised classification and quantization. For that, Kohonen's Self-Organizing Maps were used. The objective is to group the image areas into well-defined regions by reducing the smoothing of the reconstructed image, thus allowing readings to be performed that are more strongly associated with the regions of interest. EIDORS is a reference tool that forms a comparative basis for research developments on the subject of EIT. The generation of this article is part of broader studies in the field of optimization of EIT tomographic images. Studies in the area of image reconstruction using Compressed Sensing combined with Computational Intelligence techniques stand out, with the aim of reducing computational cost, increasing processing speed and increasing the quality of the reconstructed image.
Soumitra Subodh Pande, Mahesh C, GVS Ananthnath, B. Jegajothi, B. Buvaneswari, and D. Suganthi
IEEE
In 2019, the deadly COVID virus, which originated in Wuhan, China, spread throughout the entire world. It is considered to be one of the worst pandemics of this period. The epidemic had a significant influence on unemployment, and it may have even caused some fatalities. Its primary function is to assist in making judgements that are better, faster, and more trustworthy. The field of medicine is becoming an increasingly important one for the application of artificial intelligence and machine learning. This is particularly true for medical areas that make use of a wide variety of biomedical images, as well as diagnostic methods that rely on the collection and processing of a significant number of digital images. This study intended to find a way to detect anomalies in chest X-ray (CXR) images using only the most essential characteristics of deep learning. Earlier detection increases the likelihood of successful treatment; otherwise, the condition may even be fatal. This study will classify X-rays with a certain proportion of accuracy according to whether or not it is normal or abnormal. This will allow us to give the patient the appropriate treatment after viewing their X-ray. ResNet-101, DenseNet-169 and InceptionV3 are used to classify the CXR images. The ResNet-101 gives higher accuracy when compared with the other models.
Mahesh C, M. Sumithra, Madhusudana Rao Ranga, Kallakunta Ravi Kumar, D Suganthi, and S. Karthiyayini
IEEE
The exponential increase in quantity of cars on the roads of smart cities has led to traffic jams, air contamination, and delays in the delivery of essential goods. Every day road accidents continue to be a leading cause of death and serious injury. The Internet of Things-based Traffic Management System is utilized to manage traffic flow, ensure safe data transfer, and spot accidents. Autonomous vehicles and smart devices have sensors built into an IoT-based ITM system to help with identification, data collection, and transmission. Connectivity options for the Internet of Things are being severely tested by the proliferation of IoT devices and applications. This can lead to severe difficulties in the allocation of the spectrum, the resources needed to analyze the IoT data, and the allowable delay for important spare situations throughout the end-to-end communication chain. Therefore, accurate categorization or forecasting of the IoT devices' time-varying traffic characteristics is essential. However, the difficulty of making this sort of categorization remains open. Current approaches tend to be based on machine learning methods, which have a high computational cost despite not taking into account the traffic's fine-grained flow features. In this research, addresses this issue by developing a two-stage categorization system based on deep learning for IoT devices in a smart city that draws on both connectivity and statistical information. In order to excerpt the internet and statistical characteristics set for various IoT strategies, data cleaning and pre-processing should be done. Based on the results, it appears that the proposed categorization has 99% accuracy.
M. Vamsikrishna, Rajasree RS, Gopika G S, D. Suganthi, Md. Abul Ala Walid, and Ramu Kuchipudi
IEEE
Big data analytics in health service is the procedure of research in huge and different datasets, designed for uncovering hidden patterns, correlations, and trends for making the best decisions in medical field. The health and medical services are developed modern and smart healthcare platforms are made the analysis further robust for the treatment. The proper diagnosis of health records depends on primarily disease recognition and the value of accuracy is decreased if the clinical data quality is worse. But, the existing methods failed to use the learning model for handling heterogeneous healthcare information. Therefore, this article introduces Big Data Analytics with Wild Horse Optimizer based deep Learning (BDAWHO-DL) model for Healthcare Management. The proposed BDAWHO-DL technique investigates the big data in medical field and makes decisions. To do so, the BDAWHO-DL technique exploits attention based long short term memory (ABLSTM) method for data classification purposes. Moreover, the WHO system was used for the optimal hyperparameter tuning procedure of the ABLSTM algorithm to boost the classifier outcomes. The experimental outcomes indicate the promising outcomes of the BDAWHO-DL algorithm over recent approaches.
Shyamali Das, Rama Devi C, M. Gomathy Nayagam, D Suganthi, Fidha Thachamkode, and K. Manoj Senthil
IEEE
Musculoskeletal disorders cause most disability worldwide. Musculoskeletal (MSK) issues are diagnosed, predicted, and treated utilizing imaging methods, magnetic resonance imaging (MRI), and ultrasound, and their precise interpretation by specialized radiologists. Radiological scanning aids metabolic health, aging, and diabetes assessment. Proposed research shows how machine learning, specifically deep learning, may be used to analyze MRI data quickly and reliably, addressing a crucial clinical need in MSK radiology. As primary care radiography services grow, so does interest in how AI, and especially deep learning, might help radiologists and primary care physicians make accurate diagnoses and enhance patient treatment. Chronic pain impairs mobility, dexterity, and function. Diagnosis requires bone X-rays. Deep learning algorithms are increasingly used in musculoskeletal radiology, with promising outcomes. The proposed method uses a calibrated ensemble of deep learners to musculoskeletal radiograph recognition. In the suggested model uses three deep neural networks, which are commonly utilized in deep learning-based solutions for this problem domain. Publicly available dataset to demonstrates the suggested strategy outperforms three individual models and a standard ensemble learner. This study strongly supports the use of a calibrated ensemble approach for identifying abnormalities in musculoskeletal X-rays, where the greatest radiologist accuracy is between 0.95 and 0.98 in Cohen's kappa statistic.
Mohd. Asif Gandhi, K. Priya, Piyush Charan, Ritu Sharma, G. Nageswara Rao, and D. Suganthi
IEEE
Electric Vehicles (EVs) are now essential since electrifying transportation has shown to be a game-changer in raising the sustainable and eco-friendly platform in global industry. Integrating Electric Vehicle Charging System (EVCS) as a new entity into the power distribution system is one of the most important and challenging concerns. The development of an EVCS network infrastructure is a key step toward the broad adoption of EVs. In order to make informed judgments about transmission, distribution, energy allocation, and charging station placement, the control center or central aggregator must have an accurate forecast of occupancy, consumption, and energy or charging demand. Data analytics and other methods have made it possible to regularly get information from the EVCS for the purposes of archiving and processing all of the data collected. This proposed approach to presents a solution to the aforementioned problems with energy demand forecasting for EVCS networks. Three steps make up the proposed method: preprocessing, feature selection, and model performance evaluation. Preprocessing data via normalization, feature selection by K-Means, and ultimately model evaluation via K-means. The proposed model has superior results to the LSTM, GRU, and BIGRU - AM models.