Institution profile

Jahangirnagar University

Academic institutionasia · bd
Official website
Research library18linked papers
Opportunities0open roles
Selected work

Representative Papers

GeoAI-based post-segmentation quality validation of building footprints via spatial feature engineering

Aug 09, 2026

This study addresses the prevalence of topological errors in building footprints generated by deep learning models, which hinder their direct integration into GIS databases. To tackle this issue, the authors propose a multi-domain GeoAI quality control framework that fuses 24-dimensional features encompassing geometric, spatial contextual, and spectral-textural attributes. By integrating geometric regularization and spatial mutual exclusion constraints, the framework enables object-level automated quality inspection and purification. Initial masks are produced using U-Net (ResNet-34) and SAM-LoRA (ViT-B), followed by boundary deformation and duplicate object detection via decision tree classifiers. Experimental results demonstrate that the framework achieves 95.31% accuracy, 91.06% F1-score, and 0.880 Matthews correlation coefficient on an independent test area, with an 87.34% error footprint detection rate. This approach significantly enhances building database purity to 95.38% and reduces relative error by 83.09%, offering a robust and transferable quality assurance mechanism for automated GIS production.

0 citationsRead paper

Beyond Clean Text: Evaluating Encoder and Decoder Robustness for Bangla Event Detection in Noisy Text

Jun 29, 2026

This study addresses the limited robustness of current event detection systems in real-world noisy scenarios for low-resource languages such as Bengali, despite their strong performance on clean text. The authors construct a Bengali news event benchmark dataset comprising 9,979 annotated sentences spanning clean text, ASR transcripts, and spelling-perturbed variants. They systematically evaluate the noise robustness of encoder-based models (e.g., BanglaBERT, XLM-R) and instruction-tuned decoder large language models (e.g., Llama 3, Gemma 2). The work reveals a fundamental difference in noise resilience between the two architectures: decoders exhibit greater robustness when trigger words are corrupted. By integrating annotation guidelines into instruction tuning and training on mixed clean-noisy data, the performance gap is substantially narrowed. Combining model scaling with multi-source joint training yields state-of-the-art results across diverse noise conditions.

0 citationsRead paper
Recent publications

Latest Papers

GeoAI-based post-segmentation quality validation of building footprints via spatial feature engineering

Aug 09, 2026

This study addresses the prevalence of topological errors in building footprints generated by deep learning models, which hinder their direct integration into GIS databases. To tackle this issue, the authors propose a multi-domain GeoAI quality control framework that fuses 24-dimensional features encompassing geometric, spatial contextual, and spectral-textural attributes. By integrating geometric regularization and spatial mutual exclusion constraints, the framework enables object-level automated quality inspection and purification. Initial masks are produced using U-Net (ResNet-34) and SAM-LoRA (ViT-B), followed by boundary deformation and duplicate object detection via decision tree classifiers. Experimental results demonstrate that the framework achieves 95.31% accuracy, 91.06% F1-score, and 0.880 Matthews correlation coefficient on an independent test area, with an 87.34% error footprint detection rate. This approach significantly enhances building database purity to 95.38% and reduces relative error by 83.09%, offering a robust and transferable quality assurance mechanism for automated GIS production.

0 citationsRead paper

Beyond Clean Text: Evaluating Encoder and Decoder Robustness for Bangla Event Detection in Noisy Text

Jun 29, 2026

This study addresses the limited robustness of current event detection systems in real-world noisy scenarios for low-resource languages such as Bengali, despite their strong performance on clean text. The authors construct a Bengali news event benchmark dataset comprising 9,979 annotated sentences spanning clean text, ASR transcripts, and spelling-perturbed variants. They systematically evaluate the noise robustness of encoder-based models (e.g., BanglaBERT, XLM-R) and instruction-tuned decoder large language models (e.g., Llama 3, Gemma 2). The work reveals a fundamental difference in noise resilience between the two architectures: decoders exhibit greater robustness when trigger words are corrupted. By integrating annotation guidelines into instruction tuning and training on mixed clean-noisy data, the performance gap is substantially narrowed. Combining model scaling with multi-source joint training yields state-of-the-art results across diverse noise conditions.

0 citationsRead paper