Conference Papers
Permanent URI for this collectionhttps://idr.nitk.ac.in/handle/123456789/28506
Browse
8 results
Search Results
Item Offline Character recognition on Segmented Handwritten Kannada Characters(Institute of Electrical and Electronics Engineers Inc., 2019) Joe, K.G.; Savit, M.; Chandrasekaran, K.Optical character recognition (OCR) is the conversion of pictures of typed or handwritten characters into machine encoded characters. We chose to work on a subfield of OCR, namely offline learning of handwritten characters. Kannada script is agglutinative, where simple shapes are concatenated horizontally to form words. This paper presents a comparative study between different machine learning and deep learning models on Kannada characters. A Convolutional Neural Network (CNN) was chosen to show that handcrafted features are not required for recognizing classes to which characters belong to. The CNN beats the accuracy score of previous models by 5%. © 2019 IEEE.Item Hardware Accelerator for Object Detection using Tiny YOLO-v3(Institute of Electrical and Electronics Engineers Inc., 2021) Sharma, M.; Rahul, R.; Madhusudan, S.; Deepu, S.P.; Sumam David, S.For applications that require object detection to be performed in real-time, this paper presents a custom hardware accelerator, implementing state of the art Tiny YOLO-v3 algorithm. The proposed architecture achieves a reasonable tradeoff between the speed of computation (measured in frames per second or FPS) and the hardware resources required. Each CNN layer is pipelined and parameterized to make the complete design re-configurable. The proposed hardware accelerator was synthesized using the SCL(Semi-Conductor Laboratory, India) 180 nm CMOS process and also using Vivado Xilinx software with Virtex Ultrascale+ FPGA as the target device. The pipelined architecture, along with other architectural novelties, provided a higher frame-rate of 32.1 FPS and a performance of 166.4 GOPS at 200 MHz clock frequency. © 2021 IEEE.Item CNN-MFCC Model for Speaker Recognition using Emotive Speech(Institute of Electrical and Electronics Engineers Inc., 2023) Tomar, S.; Koolagudi, S.G.Finding the appropriate speaker using voice recognition is called "speaker recognition."Emotive Environment Speaker Recognition (EESR) identifies speakers using distinct emotional speech. A real-life situation that becomes a requirement for many applications is speaker recognition, which utilizes various moods. If there is no emotion in the conversation, speaker recognition algorithms work almost flawlessly. This work aims to improve the accuracy of text-dependent and emotional speaker recognition system in emotional speech contexts. The proposed method is developed using Mel-Frequency Cepstral Coefficient (MFCC) feature and the classifier considered is Convolutional Neural Networks (CNN) for various emotions. The suggested system's performance is assessed based on emotional datasets from the Kannada Language and Emotional Database (EmoDB). These emotions are present in both datasets: happy, sad, angry, fear, and neutral. Due to the complexity of emotions, speaker recognition in various emotional states is challenging. The proposed system offers an accuracy of 96.2% in the EmoDB and 97.8% in the Kannada dataset. The proposed method provides a high recognition rate for different emotions. © 2023 IEEE.Item Analysis of Speaker Recognition in Blended Emotional Environment Using Deep Learning Approaches(Springer Science and Business Media Deutschland GmbH, 2023) Tomar, S.; Koolagudi, S.G.Generally, human conversation has some emotion, and natural emotions are often blended. Today’s Speaker Recognition systems lack the component of emotion. This work proposes a Speaker Recognition approaches in Blended Emotion Environment (SRBEE) system to enhance Speaker Recognition (SR) in an emotional context. Speaker Recognition algorithms nearly always achieve perfect performance in the case of neutral speech, but it is not true from an emotional perspective. This work attempts the recognition of speakers in blended emotion with the Mel-Frequency Cepstral Coefficients (MFCC) feature extraction using the Conv2D classifier. In the blended emotional environment, calculating the accuracy of the Speaker Recognition task is complex. The blend of four basic natural emotions (happy, sad, angry, and fearful) utterances tested in the proposed system to reduce SR’s complexity in a blended emotional environment. The proposed system achieves an average accuracy of 99.3% for blended emotion with neutral speech and 92.8% for four basic blended natural emotions (happy, sad, angry, and fearful). The dataset was prepared by blending two emotions in one utterance. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.Item Hardware-Optimized Deep Learning Model for FPGA-Based Character Recognition(Institute of Electrical and Electronics Engineers Inc., 2023) Rao, P.S.; Pulikala, A.Deep neural networks (DNNs) are widely used algorithms in machine learning. Even though most of the deep learning applications are driven by software solutions, there has been significant research and development aimed at optimizing these algorithms over the years. However, when considering hardware implementation applications, it becomes essential to optimize the design not only in software but also in hardware. In this paper, we present a straightforward yet effective Convolutional Neural Network architecture that is meticulously optimized both in hardware and software for char-acter recognition applications. The implemented accelerator was realized on a Xilinx Zynq XC7Z020CLG484 FPGA using a high-level synthesis tool. To enhance performance, the accelerator employs an optimized fixed-point data type and applies loop parallelization techniques combining 2D convolution and 2D max pooling operations. The hardware efficiency of the proposed DNN is compared with some of the existing architectures in terms of hardware utilization. © 2023 IEEE.Item Convolutional Neural Network Based Approach for Automatic Detection of Diseases from Pomegranate Plants(Institute of Electrical and Electronics Engineers Inc., 2024) Rana, H.S.; Manjunatha, N.; Pokhare, S.S.; Marathe, R.A.; Rajan, J.Recent advancements in Artificial Intelligence, particularly Convolutional Neural Networks (CNNs), have significantly enhanced computer vision tasks, including plant disease detection. India is a global leader in pomegranate cultivation and production. Pomegranate is a vital horticultural commodity with significant export potential but is highly susceptible to various diseases, leading to substantial economic losses. Existing research on automatic pomegranate disease detection has limitations, as it typically analyzes either fruits or leaves in isolation and uses datasets with plain backgrounds that do not reflect real-world complexities such as lighting variations and overlapping foliage. This study proposes a novel hybrid deep learning approach that addresses these limitations. We utilized a hybrid CNN model specifically developed using the pomegranate dataset provided by the ICAR-National Research Centre on Pomegranate. This comprehensive collection includes 1,632 high-resolution images captured from orchards across India, categorized into three classes: healthy, bacterial diseases, and fungal diseases. Our model achieves an impressive accuracy rate of 98.46%, demonstrating its potential for real-world application in pomegranate disease detection and improved agricultural outcomes. © 2024 IEEE.Item Non-Invasive Detection of Anemia Using Deep Learning on Conjunctival Images(Institute of Electrical and Electronics Engineers Inc., 2025) Kedar, D.S.; Pandey, G.; Koolagudi, S.G.Anemia, characterized by low levels of red blood cells or hemoglobin, affects millions worldwide, significantly affecting public health. Traditional diagnostic methods, while effective, are invasive, costly, and inaccessible in resource-constrained settings. This paper proposes a non-invasive approach for anemia detection using conjunctival images analyzed through deep learning techniques. The proposed methodology involves capturing high-resolution conjunctival images, pre-processing them, and using a customized Convolutional Neural Network (CNN) for feature extraction and classification. The results achieved by the customized CNN fine-tuned with a batch size of 16 give an Accuracy of 96%, Precision of 95%, Recall of 96%, and ROC-AUC score of 0.99. The customized CNN outperformed the other models for this work, such as Random Forest, XGBoost, SVM, ResNet50, and MobileNetV2. This work highlights the potential for non-invasive diagnostic tools to improve accessibility and efficiency in healthcare, particularly for underserved populations. The findings endorse integrating deep learning in healthcare as a transformative approach to address global challenges such as anemia. © 2025 IEEE.Item Machine Learning-Based Comprehensive Space Weather Monitoring and Prediction System(American Institute of Aeronautics and Astronautics Inc, AIAA, 2025) Israni, K.; Iyer, E.J.A.; Paul, N.Space weather events, driven by solar phenomena such as solar flares, coronal mass ejections (CMEs), and solar winds, present critical challenges to both space-based and terrestrial systems. These phenomena can cause geomagnetic storms that disrupt satellite communications, degrade GPS accuracy, overload power grids, and impair aviation and communication networks. Understanding the mechanisms and impacts of space weather is crucial for developing predictive capabilities to mitigate these risks. This study investigates the physical drivers of space weather and introduces a machine learning (ML)-driven framework for real-time monitoring and prediction. Leveraging multi-source data from satellites and ground-based observatories, the system incorporates advanced sunspot classification models based on the McIntosh and Wilson methods. By analyzing solar images and magnetograms with convolutional and recurrent neural networks, the framework achieves significant improvements in prediction accuracy for solar flares and geomagnetic disturbances. Furthermore, the integration of data-driven technologies with traditional observation methods provides timely and actionable forecasts, enhancing resilience against the cascading effects of space weather disruptions on global infrastructure. This research sets a foundation for scalable solutions to safeguard critical systems and supports proactive decision-making in space weather preparedness. © 2025, American Institute of Aeronautics and Astronautics Inc, AIAA. All rights reserved.