http://https://dspace.iiti.ac.in:80 The DSpace digital repository system captures, stores, indexes, preserves, and distributes digital research material. 2026-10-15T13:23:59Z https://dspace.iiti.ac.in/handle/123456789/19441 Title: Unsupervised clustering of tear damage in soybean kernels Authors: Miglani, Ankur (55578386900); Kankar, Pavan Kumar (8367238500) Abstract: Tear damage in soybean kernels is a significant post-harvest defect that reduces seed quality. In this study, a fully automated, label-free machine vision pipeline is developed for characterising tear damage in soybean kernels. A dataset consisting of 1412 high-magnification (4.5x) images collected from seven soybean cultivars was developed, covering a diverse spectrum of tear morphologies ranging from short abrasive scratches to large-scale axial and lateral cuts in the seed coat. To isolate the damaged regions, a subject-centric preprocessing workflow is developed that standardises illumination and enhances defect contrast. Next, we applied Segment Anything Model 2 (SAM2) to generate high-recall instance masks. Subsequently, these segmented regions are encoded using a MobileNetV3-based cluster-guided contrastive representation learning (CRL) framework. This framework incorporates clustering-driven sampling and reweighted NT-Xent training to produce discriminative 128-dimensional embeddings that preserve subtle, fine-scale textural and structural information. Next, different unsupervised clustering algorithms, namely K-Means, GMM, Agglomerative (ward, average, and complete), Spectral, DBSCAN, and HDBSCAN, are evaluated using Silhouette, Davies-Bouldin (DB), and Calinski-Harabasz (CH) scores. Agglomerative (ward) was identified as the best-performing spatial method. To establish a reproducible, open-world strategy, the optimal number of clusters () was determined as a data-driven outcome through a dual-validation approach, maximizing hierarchical stability (Silhouette sweep) and corroborated by probabilistic model selection (BIC/AIC). A blind expert validation using 290 images further confirms strong semantic alignment between consensus expert annotations and cluster-based labels, yielding a Cohen’s kappa of 0.868 with a 95% confidence interval of 0.826-0.910. The analysis of 1687 damaged regions segmented from these 8 damage classes reveals that each damage type exhibits substantial intra-class variability, multi-scale damage distributions, and centroid concentrations near central kernel zones. The proposed pipeline establishes a highly precise, annotation-efficient foundation for detailed morphological profiling and objective quality assessment under controlled imaging conditions, advancing the paradigm away from subjective manual inspection. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2026. 2026-01-01T00:00:00Z https://dspace.iiti.ac.in/handle/123456789/19447 Title: Isolation and characterization of multitrait Bacillus spp. that exhibit in vitro bioremediation of heavy metals and pesticides Authors: Mandpe, Ashootosh (57200034951) Abstract: Soil health and agricultural productivity are severely threatened by agrochemical pollution and heavy metal accumulation. This study aimed to isolate, characterize, and construct a multi-strain Bacillus spp. consortium possessing simultaneous plant growth-promoting rhizobacteria (PGPR) traits, heavy metal tolerance, and pesticide degradation capabilities to serve as a sustainable bioremediation tool. Spore-forming bacteria were isolated from 44 diverse Indian crop rhizosphere soil samples via enrichment and heat-shock treatment. Non-hemolytic isolates were screened for mineral solubilization, diazotrophic growth, and inter-strain compatibility. Formulated multi-strain blends were evaluated for heavy metal tolerance against HgCl2, Pb(CH3COO)2, NiCl2, and As2O3 and compatibility with six agrochemicals. Malathion biodegradation was tracked using plate assays and confirmed via liquid chromatography-tandem mass spectrometry (LC–MS/MS). Six distinct Bacillus strains (Bacillus amyloliquefaciens, Bacillus subtilis, Bacillus haynesii, Bacillus tequilensis, Bacillus licheniformis, and Bacillus halotolerans) were identified. These strains demonstrated specialized phosphate (up to 65.13 mg/L) and potash (up to 9.43 mg/L) solubilization. Excluding the antagonistic B. tequilensis, four consortia were developed. Blend 2 (B. amyloliquefaciens–B. subtilis–B. licheniformis) exhibited the highest heavy metal resilience, tolerating lead up to 1000 mg/L. Crucially, B. haynesii and B. licheniformis actively degraded the organophosphate malathion. LC–MS/MS verified biotransformation, mapping the metabolic breakdown into malathion monocarboxylic acid (MMC), and malathion dicarboxylic acid (MDC). This study presents the first official report of breakdown of malathion by B. haynesii. The developed multi-strain Bacillus consortia show promising functional advantages over single-strain inoculants, providing a highly effective, multifaceted strategy for sustainable agriculture and environmental bioremediation. © 2026 Taylor & Francis Group, LLC. 2026-01-01T00:00:00Z https://dspace.iiti.ac.in/handle/123456789/19442 Title: The design of a multi-band antenna with improved higher-order mode radiation using characteristic mode analysis for L5-band, L1-band, and S-band applications Authors: Joshi, Prachi (58688424700) Abstract: A novel multi-band antenna with an improved higher-order mode radiation pattern based on Characteristic Mode Analysis (CMA) is presented. The tri-band characteristic is achieved by exciting one higher-order mode and two orthogonal modes. The initial orthogonal mode is achieved by converting the antenna from a circular to an elliptical shape. Surface current reshaping transforms the higher-order conical radiation pattern mode into a broadside direction. The surface current nulls with out-of-phase are moved toward the patch corners using CMA. Characteristic modes are excited with a coaxial feed, which uses full-wave Electromagnetic (EM) simulation. It excites two orthogonal modes at 1176 MHz and 1575 MHz for L5-band and L1-band applications, respectively, and one higher-order mode at 2500 MHz for S-band applications. The proposed antenna has moderate gain and broadside radiation patterns across the operational frequency bands. The working of the antenna and validation are represented by equivalent circuit modeling. The experimental results present excellent agreement with EM simulated results. © 2025 Sharif University of Technology. 2025-01-01T00:00:00Z https://dspace.iiti.ac.in/handle/123456789/19445 Title: Ni-Loaded SnO2-MoS2 Heterostructure for Highly Selective and Low-Temperature SO2 Sensing With Real-Time Device Integration Authors: Verma, Vikash Kumar (60570042100); Mukherjee, Sourajit (60111482700); Bhowmick, Shorinjiryu (60111927300); Phogat, Usha (60872296800); Mukherjee, Shaibal (58302491700) Abstract: Sulfur dioxide (SO2) is a critical pollutant associated with severe respiratory and cardiovascular effects, necessitating continuous monitoring in industrial and environmental settings. Existing analytical systems are accurate but bulky, expensive, and unsuitable for portable real-time operation. Current metal-oxide and 2D material-based sensors still face challenges including limited selectivity, high operating temperatures, and slow sensing kinetics. This work aims to develop a highly selective SO2 sensor using nickel-loaded SnO2-MoS2 hybrid nanostructures integrated on interdigitated electrode platforms. Pristine MoS2 nanoflowers were synthesized, hybridized with SnO2, and loaded with controlled Ni concentrations to enhance electronic modulation and surface adsorption. The fabricated films were characterized using FESEM, EDS, Raman, XRD, and UV-Vis and tested under dynamic gas exposure with real-time resistance acquisition. The 4N-SM-4 sample (4:1 SnO2: MoS2 with 4% Ni) demonstrated the best sensing behavior, achieving a response of 1609% at 20 ppm with a response time of 38 s, recovery time of 18 s, and a selectivity coefficient of 7. The limit of detection and quantification were 0.144 and 0.604 ppm, respectively. Excellent stability was observed across varying humidity and during 40-day continuous evaluation. The portable IDE-based readout system enables practical real-time deployment. © 2026 Wiley-VCH GmbH. 2026-01-01T00:00:00Z