<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:dc="http://purl.org/dc/elements/1.1/" version="2.0">
  <channel>
    <title>DSpace Community:</title>
    <link>http://localhost:8081/jspui/handle/123456789/15072</link>
    <description />
    <pubDate>Thu, 19 Nov 2026 20:26:12 GMT</pubDate>
    <dc:date>2026-11-19T20:26:12Z</dc:date>
    <item>
      <title>DEVELOPMENT OF DEEP LEARNING FRAMEWORKS FOR SEED CATEGORIZATION VIA NIR-HYPERSPECTRAL IMAGING</title>
      <link>http://localhost:8081/jspui/handle/123456789/21775</link>
      <description>Title: DEVELOPMENT OF DEEP LEARNING FRAMEWORKS FOR SEED CATEGORIZATION VIA NIR-HYPERSPECTRAL IMAGING
Authors: Tyagi, Nitin
Abstract: The need for computational systems capable of accurately classifying and analyzing&#xD;
wheat seeds using Machine Learning (ML) and Hyperspectral Imaging (HSI) has grown&#xD;
significantly in recent years. HSI provides rich spectral information, making it a powerful&#xD;
tool for nondestructive seed classification in agricultural and industrial applications.&#xD;
This thesis presents a comprehensive study on wheat seed classification by integrating&#xD;
Deep Learning (DL) techniques with Near-Infrared Hyperspectral Imaging (NIR-HSI).&#xD;
The research aims to enhance seed identification’s accuracy, scalability, and practical&#xD;
applicability in agricultural and industrial settings.&#xD;
The research begins by exploring wheat species and varieties, their origins, significance,&#xD;
and classification techniques, ranging from traditional methods (visual inspection,&#xD;
chemical analysis) to modern nondestructive approaches, including near-infrared spectroscopy&#xD;
and HSI. A detailed literature review discusses ML and DL techniques in seed&#xD;
classification, including hyperspectral and RGB imaging, multimodal fusion, spectral&#xD;
band selection, and preprocessing methods for spectral enhancement and evaluation&#xD;
metrics. From the literature, it was found that a key challenge in seed classification&#xD;
is the lack of publicly available seed datasets. To address this, three distinct wheat&#xD;
seed datasets are prepared, including a single-seed hyperspectral dataset, a multimodal&#xD;
dataset captured using both RGB and HSI systems, and a bulk seed dataset simulating&#xD;
real-world industrial applications.</description>
      <pubDate>Sat, 01 Feb 2025 00:00:00 GMT</pubDate>
      <guid isPermaLink="false">http://localhost:8081/jspui/handle/123456789/21775</guid>
      <dc:date>2025-02-01T00:00:00Z</dc:date>
    </item>
    <item>
      <title>UTILITY ENHANCEMENT AND BIAS MITIGATION IN SKIN LESION ANALYSIS USING DEEP LEARNING</title>
      <link>http://localhost:8081/jspui/handle/123456789/21774</link>
      <description>Title: UTILITY ENHANCEMENT AND BIAS MITIGATION IN SKIN LESION ANALYSIS USING DEEP LEARNING
Authors: Pundhir, Anshul
Abstract: This thesis provides an in-depth investigation into integrating deep learning techniques&#xD;
in medical imaging, a field undergoing rapid evolution due to technological&#xD;
advancements. The main objective is to improve the utility and fairness of deep learning&#xD;
models in the context of medical imaging, thereby bridging the gap between computational&#xD;
algorithms and real-world medical applications. The research covers various&#xD;
crucial areas, including medical image classification, multimodal fusion, enhanced feature&#xD;
representations, explainability of deep learning models, and mitigating bias to&#xD;
ensure fairness in medical diagnosis. This research endeavors to confront and mitigate&#xD;
essential challenges within medical imaging by introducing novel methodologies across&#xD;
various fronts.&#xD;
We propose multimodal fusion approaches to improve diagnostic accuracy in skin&#xD;
lesion classification by seamlessly integrating information from diverse data sources.&#xD;
These approaches hold promise in enhancing decision-making processes and improving&#xD;
overall diagnostic outcomes by harnessing the complementary nature of different modalities,&#xD;
such as skin images and patients’ demographics. Our study also delves into utilizing&#xD;
visual attention mechanisms within convolutional neural networks to enhance feature&#xD;
representation through context supervision. By selectively highlighting salient regions&#xD;
in medical images and incorporating global and visual context representations, our&#xD;
methodology aims to yield more informative features, augmenting model performance&#xD;
and contributing to more accurate diagnostic assessments.</description>
      <pubDate>Mon, 01 Jul 2024 00:00:00 GMT</pubDate>
      <guid isPermaLink="false">http://localhost:8081/jspui/handle/123456789/21774</guid>
      <dc:date>2024-07-01T00:00:00Z</dc:date>
    </item>
    <item>
      <title>SCENE TEXT IMAGE SUPER-RESOLUTION USING DEEP LEARNING</title>
      <link>http://localhost:8081/jspui/handle/123456789/21756</link>
      <description>Title: SCENE TEXT IMAGE SUPER-RESOLUTION USING DEEP LEARNING
Authors: Singh, Shrey
Abstract: Scene Text Image Super-Resolution (STISR) is a vital sub-task in document analysis&#xD;
and optical character recognition (OCR), particularly in real-world applications such&#xD;
as license plate recognition, translation, assistive technologies, and augmented reality.&#xD;
In unconstrained environments, text captured in natural scenes often su!ers from&#xD;
motion blur, low resolution, defocus, and other degradations, making it di"cult for&#xD;
both humans and machines to interpret the content accurately. Traditional superresolution&#xD;
techniques, primarily designed for natural images, often fail to preserve&#xD;
the fine-grained semantic and structural features of text, such as strokes, curves, and&#xD;
character boundaries. This thesis addresses these challenges by proposing a suite of&#xD;
di!usion-based generative models that incorporate structural, semantic, and linguistic&#xD;
priors to restore high-quality, readable text images.&#xD;
The first contribution introduces the Di!usion-Conditioned Di!usion Model (DCDM),&#xD;
which integrates low-resolution visual input with character-level semantic embeddings&#xD;
derived from a latent text di!usion module. The second contribution presents the&#xD;
Skeleton-Aware Di!usion Model (SADM), where refined skeleton maps act as structural&#xD;
guides during the di!usion process to improve character-level fidelity. The third and&#xD;
final contribution explores on the stroke level by the Stroke-Conditioned Di!usion&#xD;
Model (SCDM), focusing on low-resource Indic scripts such as Hindi, Marathi, and&#xD;
Gujarati. SCDM leverages a novel stroke decomposition strategy to generalize across&#xD;
linguistically similar scripts through structural supervision.</description>
      <pubDate>Tue, 01 Apr 2025 00:00:00 GMT</pubDate>
      <guid isPermaLink="false">http://localhost:8081/jspui/handle/123456789/21756</guid>
      <dc:date>2025-04-01T00:00:00Z</dc:date>
    </item>
    <item>
      <title>Enhancing Few-shot Semantic Segmentation Technique for Medical Image Segmentation</title>
      <link>http://localhost:8081/jspui/handle/123456789/21597</link>
      <description>Title: Enhancing Few-shot Semantic Segmentation Technique for Medical Image Segmentation
Authors: Bagde, Abhishek
Abstract: When confronted with new, unseen semantic classes, fully-supervised deep learning seg&#xD;
mentation models suffer, as their fine-tuning necessitates enormous quantities of data with&#xD;
annotation. Few-shot semantic segmentation (FSS) offers a solution by enabling the seg&#xD;
mentation of previously unseen semantic classes by using only a few labelled examples&#xD;
and without any reliance on fine-tuning. Existing state-of-the-art FSS approaches primar&#xD;
ily focus on segmenting natural imageset and rely on abundant annotated training data for&#xD;
effective generalization to testing classes that are unseen. Due to the above reasons, this&#xD;
training technique is not practical in clinical imaging situations with scarce annotations. To&#xD;
overcome the difficulties mentioned above, our work proposes enhancement in the exist&#xD;
ing self-supervised FSS framework for medical images called SSL-ALPNet; this framework&#xD;
eliminates the need for annotations which were required in the training phase and leverages&#xD;
superpixel-based pseudo-labels that provide supervision signals and also includes an adap&#xD;
tive local prototype pooling module integrated into the prototypical networks for improvis&#xD;
ing segmentation accuracy. The efficacy ofourmodelisdemonstratedinorgansegmentation&#xD;
using Abdominal-MRI images. Experimental results show that our model, which is an en&#xD;
hancement made to the SSL-ALPNet [1] approach, achieves better segmentation accuracy as&#xD;
compared to other standard Few-Shot-Segmentation methods whose training is based on a&#xD;
manually annotated dataset.</description>
      <pubDate>Thu, 01 Jun 2023 00:00:00 GMT</pubDate>
      <guid isPermaLink="false">http://localhost:8081/jspui/handle/123456789/21597</guid>
      <dc:date>2023-06-01T00:00:00Z</dc:date>
    </item>
  </channel>
</rss>

