Saeed Sharifian
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Saeed Sharifian has a total of 63 co-authors. Here are the top co-authors by number of shared publications:
- Seyed Ahmad Motamedi
Electrical Engineering Department, Amirkabir University of Technology, Tehran, Iran
16 shared publications Hassan Taheri
Department of Electrical and Electronic, Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran
4 shared publicationsMina Emami Khansari
4 shared publications- Ahmad Rahati Belabad
Amirkabir University of Technology
3 shared publications - Alireza Farshin
NVIDIA, Stockholm, Sweden
3 shared publications Mohammad Kazem Akbari
Computer Engineering & Information Technology Department, Amirkabir University of Technology, Tehran, Iran
3 shared publications
Latest Publications
- Open Access
- 13 August 2026
- Published by Elsevier in SSRN Electronic Journal
CFGPNet: Cross-Attention-Based Fused Gradient Programmed Network Framework for Multispectral Object Detection
RGB--T object detection exploits the complementary strengths of visible and infrared imagery, providing a robust solution for perception in low-light, adverse-weather, and complex multi-scale environments. However, existing methods still suffer from several limitations, including insufficient cross-modal interaction, unstable fusion caused by the distribution gap between modalities, and the high computational cost of heavy attention-based architectures. To address these issues, CFGPNet is proposed, a Cross-Attention-Based Fused Gradient Programmed Network framework for multispectral object detection. CFGPNet builds on an improved GELAN backbone that uses RepViT-style re-parameterized blocks. This design strengthens feature representation while preserving computational efficiency. In addition, a Cross Computation Efficient Attention (CrossCEA) module is introduced to enhance cross-modal feature interaction and reduce redundant information transfer between visible and thermal branches. To generate compact and discriminative fused representations, an Attention Selection and Aggregation Fusion (ASAF) network is further designed to combine dense feature aggregation with selective attention-based emphasis. Moreover, a programmable-gradient auxiliary branch is integrated into each CFGPNet variant to improve gradient delivery and optimization quality. Extensive experiments on five public multispectral benchmarks, FLIR, M3FD, LLVIP, VEDAI, and MFAD, demonstrate that CFGPNet achieves strong and consistent performance across diverse scene types, object scales, and modality balances. In particular, the proposed framework attains 80.7% mAP50 / 45.0% mAP50:95 on FLIR, 89.9% / 63.4% on M3FD, and 97.8% / 68.9% on LLVIP. It also reaches 83.3% / 56.9% on VEDAI and 83.4% / 61.8% on MFAD. These results show that CFGPNet provides an effective and practical solution for multispectral object detection while offering useful accuracy--efficiency trade-offs through its three model scales. The code, data, and fine-tuned models are available at https://github.com/NimaHatami99/CFGPNet.
- 13 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessCFGPNet: Cross-Attention-Based Fused Gradient Programmed Network Framework for Multispectral Object Detection
RGB--T object detection exploits the complementary strengths of visible and infrared imagery, providing a robust solution for perception in low-light, adverse-weather, and complex multi-scale environments. However, existing methods still suffer from several limitations, including insufficient cross-modal interaction, unstable fusion caused by the distribution gap between modalities, and the high computational cost of heavy attention-based architectures. To address these issues, CFGPNet is proposed, a Cross-Attention-Based Fused Gradient Programmed Network framework for multispectral object detection. CFGPNet builds on an improved GELAN backbone that uses RepViT-style re-parameterized blocks. This design strengthens feature representation while preserving computational efficiency. In addition, a Cross Computation Efficient Attention (CrossCEA) module is introduced to enhance cross-modal feature interaction and reduce redundant information transfer between visible and thermal branches. To generate compact and discriminative fused representations, an Attention Selection and Aggregation Fusion (ASAF) network is further designed to combine dense feature aggregation with selective attention-based emphasis. Moreover, a programmable-gradient auxiliary branch is integrated into each CFGPNet variant to improve gradient delivery and optimization quality. Extensive experiments on five public multispectral benchmarks, FLIR, M3FD, LLVIP, VEDAI, and MFAD, demonstrate that CFGPNet achieves strong and consistent performance across diverse scene types, object scales, and modality balances. In particular, the proposed framework attains 80.7% mAP50 / 45.0% mAP50:95 on FLIR, 89.9% / 63.4% on M3FD, and 97.8% / 68.9% on LLVIP. It also reaches 83.3% / 56.9% on VEDAI and 83.4% / 61.8% on MFAD. These results show that CFGPNet provides an effective and practical solution for multispectral object detection while offering useful accuracy--efficiency trade-offs through its three model scales. The code, data, and fine-tuned models are available at https://github.com/NimaHatami99/CFGPNet.
- Open Access
- 10 June 2026
- Published by SCI AND TECH UNIVERSAL INC in Scientia Iranica
- p. e24236
Stochastic Attention Refinement for Remote Sensing Change Detection: Learning Adaptive Modulation Patterns Through Contextual Pattern Embedding
Accurate change detection in remote sensing imagery requires sophisticated multi-scale temporal feature integration and attention mechanisms. Current methods suffer from suboptimal multi-scale information utilization due to uniform attention deployment and insufficient feature discriminability from deterministic processing. We propose a novel framework addressing these limitations through two key innovations. First, an Adaptive Scale-Context Attention module with resolution-aware orchestration strategically applies spatial attention at higher scales for precise boundary delineation and channel attention at lower scales for semantic feature selection. Second, and most importantly, we introduce a stochastic attention refinement mechanism that revolutionizes attention-based change detection by learning adaptive modulation patterns through contextual pattern embedding. This stochastic framework employs posterior and prior distributions to model context-dependent enhancement patterns, applying learned contextual representations to dynamically calibrate attention scores and significantly improve feature discriminability beyond deterministic approaches. Our method processes bi-temporal images through dual-stream encoders, applies Adaptive Scale-Context Attention modules with stochastic enhancement across multiple scales, and reconstructs change maps through semantically-aware upsampling. Extensive experiments on four benchmark datasets demonstrate superior performance: we achieve 93.82% F1-score on DSIFN-CD, 94.37% F1-score on WHU building dataset, 91.85% F1-score on LEVIR-CD, and 76.19% F1-score on MSRS-CD, while maintaining computational efficiency with only 6.8M parameters. Comprehensive ablation studies validate the effectiveness of both resolution-aware attention orchestration and stochastic enhancement, establishing a new paradigm for efficient and accurate remote sensing change detection.
- 10 June 2026
- Published by SCI AND TECH UNIVERSAL INC in Scientia Iranica
- p. e24236
Open AccessStochastic Attention Refinement for Remote Sensing Change Detection: Learning Adaptive Modulation Patterns Through Contextual Pattern Embedding
Accurate change detection in remote sensing imagery requires sophisticated multi-scale temporal feature integration and attention mechanisms. Current methods suffer from suboptimal multi-scale information utilization due to uniform attention deployment and insufficient feature discriminability from deterministic processing. We propose a novel framework addressing these limitations through two key innovations. First, an Adaptive Scale-Context Attention module with resolution-aware orchestration strategically applies spatial attention at higher scales for precise boundary delineation and channel attention at lower scales for semantic feature selection. Second, and most importantly, we introduce a stochastic attention refinement mechanism that revolutionizes attention-based change detection by learning adaptive modulation patterns through contextual pattern embedding. This stochastic framework employs posterior and prior distributions to model context-dependent enhancement patterns, applying learned contextual representations to dynamically calibrate attention scores and significantly improve feature discriminability beyond deterministic approaches. Our method processes bi-temporal images through dual-stream encoders, applies Adaptive Scale-Context Attention modules with stochastic enhancement across multiple scales, and reconstructs change maps through semantically-aware upsampling. Extensive experiments on four benchmark datasets demonstrate superior performance: we achieve 93.82% F1-score on DSIFN-CD, 94.37% F1-score on WHU building dataset, 91.85% F1-score on LEVIR-CD, and 76.19% F1-score on MSRS-CD, while maintaining computational efficiency with only 6.8M parameters. Comprehensive ablation studies validate the effectiveness of both resolution-aware attention orchestration and stochastic enhancement, establishing a new paradigm for efficient and accurate remote sensing change detection.
- Open Access
- 3 February 2026
- Published by Elsevier in SSRN Electronic Journal
Multi-Teacher Knowledge Distillation for Efficient Satellite-Based Change Detection in Remote Sensing: Advancing Edge Computing for Real-Time Environmental Hazard Monitoring in LEO Networks
The rapid growth of low Earth orbit satellite constellations has revolutionized Earth observation, yet processing high-resolution remote sensing data on resource-constrained hardware remains challenging for time-sensitive natural disaster management applications. We introduce the Multi-Teacher Knowledge Distillation Change Detection (MTKDCD) framework, which leverages a Kubernetes orchestrated distributed edge computing system model to enable distributed AI intelligence across LEO satellite networks, achieving low-latency, resilient on-board change detection for real-time responses to hazards such as earthquakes, floods, and wildfires. The teacher model features a Contextual Adaptive Dual Attention Fusion module with linear self-attention for capturing global and local details in imagery. A multi-teacher ensemble, trained sequentially with bootstrap sampling and Jensen-Shannon divergence-based diversity loss on resource-constrained worker nodes and distilled in parallel over inter-satellite links, transfers robust knowledge to a lightweight student model suited for distributed on-board deployment with local reference feature caching. A compound loss combining Binary Cross-Entropy, Generalized Dice Loss, and novel Boundary Density Loss enhances edge delineation and robustness to annotation imperfections in remote sensing data. The distilled student achieves competitive accuracy with 1.51M parameters and 3.60B FLOPs, reducing parameters by 95.41% and complexity by 90.95% versus the ensemble teacher. Experiments on three benchmarks show teacher F1-scores of 92.00%, 93.15%, and 97.13%, with ablations confirming effectiveness for real-time hazard monitoring.
- 3 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessMulti-Teacher Knowledge Distillation for Efficient Satellite-Based Change Detection in Remote Sensing: Advancing Edge Computing for Real-Time Environmental Hazard Monitoring in LEO Networks
The rapid growth of low Earth orbit satellite constellations has revolutionized Earth observation, yet processing high-resolution remote sensing data on resource-constrained hardware remains challenging for time-sensitive natural disaster management applications. We introduce the Multi-Teacher Knowledge Distillation Change Detection (MTKDCD) framework, which leverages a Kubernetes orchestrated distributed edge computing system model to enable distributed AI intelligence across LEO satellite networks, achieving low-latency, resilient on-board change detection for real-time responses to hazards such as earthquakes, floods, and wildfires. The teacher model features a Contextual Adaptive Dual Attention Fusion module with linear self-attention for capturing global and local details in imagery. A multi-teacher ensemble, trained sequentially with bootstrap sampling and Jensen-Shannon divergence-based diversity loss on resource-constrained worker nodes and distilled in parallel over inter-satellite links, transfers robust knowledge to a lightweight student model suited for distributed on-board deployment with local reference feature caching. A compound loss combining Binary Cross-Entropy, Generalized Dice Loss, and novel Boundary Density Loss enhances edge delineation and robustness to annotation imperfections in remote sensing data. The distilled student achieves competitive accuracy with 1.51M parameters and 3.60B FLOPs, reducing parameters by 95.41% and complexity by 90.95% versus the ensemble teacher. Experiments on three benchmarks show teacher F1-scores of 92.00%, 93.15%, and 97.13%, with ablations confirming effectiveness for real-time hazard monitoring.
- 21 February 2025
- Published by Springer Nature in Signal, Image and Video Processing
- Vol. 19 (4) , 1-9
Explainable AI enhanced transformer based UNet for medical images segmentation using gradient weighted class activation map
- 21 February 2025
- Published by Springer Nature in Signal, Image and Video Processing
- Vol. 19 (4) , 1-9
Explainable AI enhanced transformer based UNet for medical images segmentation using gradient weighted class activation map