Region-Aware Degradation-Conditioned Vision-Language Learning for Blind Image-Quality Assessment of Power Field-Operation Imagery
- 1Electric Power Research Institute of Guizhou Power Grid Co., Ltd., Guiyang 550002, China
- 2School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
Abstract
Reliable blind image-quality assessment is essential for visual monitoring of power-field operations, where images captured by fixed cameras, unmanned aerial vehicles, and handheld terminals are frequently affected by blur, noise, uneven exposure, compression, and resolution loss. These degradations are often spatially non-uniform, and distortions occurring in task-relevant regions usually have a greater impact on operation-safety assessment than those appearing in background regions. Existing methods, however, commonly rely on global representations or treat local regions uniformly, making it difficult to identify degradations in task-critical areas and generalize to complex industrial imagery. To address this problem, we propose RDC-IQA, a region-aware degradation-conditioned vision-language framework. It combines hierarchical visual tokens with contrastively learned degradation priors through confidence-aware residual conditioning, preserving semantic representations while improving sensitivity to local distortions. The resulting regional features are aligned with structured descriptions of scene content, distortion attributes, and quality levels. Regional contributions are then determined jointly by task relevance and degradation severity, so that impaired task-relevant regions receive greater attention during quality prediction. In addition, a local degradation ranking objective learns spatial quality variations without requiring patch-level subjective annotations. Experiments on six public benchmarks and the PowerIQA dataset demonstrate that RDC-IQA achieves strong performance across synthetic, authentic, and power field-operation imagery.
Related
- Toward Controlling Biology with Language:Offline Learning of Prompt-Conditioned Interventions for Cells, Organoids, and Biobots
- Simultaneous multi-task assessment of five neurological language disorders via cross-corpus multimodal learning
- Protein Language Model-Conditioned Graph Neural Networks for Multitask GPCR Ligand Activity Prediction
- Toward an Objective Assessment of Fight-or-Flight System Function in Anxiety Disorders: Startle Reflex Modulation in Threat Coping Contexts as a Candidate Neurophysiological Marker of Disrupted Defense System Operation
- MIEN: a deep-learning model for pocket-conditioned glycan pose and conformation discrimination
BibTeX
@article{meng2026regionaware,
title = {{Region-Aware Degradation-Conditioned Vision-Language Learning for Blind Image-Quality Assessment of Power Field-Operation Imagery}},
author = {Lingwen Meng and Guobang Ban and Jintong Ma and Siqi Guo and Xinshan Zhu},
journal = {Applied Sciences},
year = {2026},
month = oct,
doi = {10.3390/app162010374},
url = {https://doi.org/10.3390/app162010374},
note = {CC BY},
}
RIS
TY - JOUR
TI - Region-Aware Degradation-Conditioned Vision-Language Learning for Blind Image-Quality Assessment of Power Field-Operation Imagery
AU - Lingwen Meng
AU - Guobang Ban
AU - Jintong Ma
AU - Siqi Guo
AU - Xinshan Zhu
PY - 2026
DA - 2026/10/26
JO - Applied Sciences
DO - 10.3390/app162010374
UR - https://doi.org/10.3390/app162010374
AB - Reliable blind image-quality assessment is essential for visual monitoring of power-field operations, where images captured by fixed cameras, unmanned aerial vehicles, and handheld terminals are frequently affected by blur, noise, uneven exposure, compression, and resolution loss. These degradations are often spatially non-uniform, and distortions occurring in task-relevant regions usually have a greater impact on operation-safety assessment than those appearing in background regions. Existing methods, however, commonly rely on global representations or treat local regions uniformly, making it difficult to identify degradations in task-critical areas and generalize to complex industrial imagery. To address this problem, we propose RDC-IQA, a region-aware degradation-conditioned vision-language framework. It combines hierarchical visual tokens with contrastively learned degradation priors through confidence-aware residual conditioning, preserving semantic representations while improving sensitivity to local distortions. The resulting regional features are aligned with structured descriptions of scene content, distortion attributes, and quality levels. Regional contributions are then determined jointly by task relevance and degradation severity, so that impaired task-relevant regions receive greater attention during quality prediction. In addition, a local degradation ranking objective learns spatial quality variations without requiring patch-level subjective annotations. Experiments on six public benchmarks and the PowerIQA dataset demonstrate that RDC-IQA achieves strong performance across synthetic, authentic, and power field-operation imagery.
SN - 2076-3417
ER -