Dual-view X-ray luggage screening remains challenging because occlusion, projection overlap, fragmented object evidence, and uneven category difficulty jointly weaken image-level multi-label recognition. Existing dual-view methods mainly enhance cross-view representation, while the coupling between training-stage learning-instability estimation and high-resolution feature modulation remains less explored. To address this issue, we propose a Modulated Class-Balancing Unified framework, termed MCB-U, which integrates Modulated Class-Balancing Dynamic Weighting with CoordAtt-U. MCB-DW estimates class-level learning instability from intra-batch prediction dispersion and maintains an EMA-smoothed class-weight signal, while CoordAtt-U uses the detached signal to modulate the high-resolution N3 attention response. On the official DvXray classification protocol over three random seeds, MCB-U achieves an mAP of 0.9278 ± 0.0104, improving the Binary Cross-Entropy baseline by 0.0370 mAP with marginal computational overhead. Modulation-path ablations further verify the effectiveness of the dynamic gain design, and per-class analysis shows consistent improvements on evaluation-level hard categories.

Since the Industrial Revolution, anthropogenic activities, primarily fossil fuel combustion, have driven a sharp increase in ${\mathrm{CO}}_2$ emissions, making them the principal driver of global climate change. Precise monitoring and quantification of ${\mathrm{CO}}_2$ emissions are essential for effective greenhouse gas mitigation. Traditional “bottom-up” inventories often suffer from limited timeliness, low spatial resolution, and significant uncertainties. Satellite remote sensing offers an alternative “top-down” approach for emission estimation. Compared to existing ${\mathrm{CO}}_2$ sensors, ${\mathrm{NO}}_2$ observation satellites provide higher spatiotemporal resolution. Given that ${\mathrm{NO}}_2$ and ${\mathrm{CO}}_2$ are co-emitted during combustion with a stable relationship, ${\mathrm{NO}}_2$ can serve as an effective proxy to indirectly derive ${\mathrm{CO}}_2$ emissions. In this study, an exploratory framework for city-scale ${\mathrm{CO}}_2$ estimation was developed using TROPOMI ${\mathrm{NO}}_2$ column concentrations, MERRA-2 wind fields, EDGAR inventory and the ODIAC inventory. The analysis focused on seven major cities in Shandong Province, China, from April to September 2022. By integrating a wind-rotation technique with a line density model and the differential evolution (DE) algorithm, we derived ${\mathrm{NO}}_2$ emissions and atmospheric lifetimes. The ${\mathrm{NO}}_2{\mathrm{-to-CO}}_2$ relationship was established based on sector-weighted inventory data to quantify fossil-fuel ${\mathrm{CO}}_2$ fluxes. The results identify Qingdao, Jinan, and Linyi as emission hotspots, followed by Rizhao, with lower emissions observed in Yantai, Liaocheng, and Jining. Comparison with the ODIAC inventory illustrates that this framework provides a top-down constraint for identifying localized emission characteristics and potential discrepancies in bottom-up datasets. This study offers a complementary tool for near-real-time urban carbon monitoring during the non-heating season.