A new capture vaporizer (CV) has been developed and used recently in the Aerodyne aerosol mass spectrometer (AMS) and aerosol chemical speciation monitor (ACSM) instead of the standard vaporizer (SV) to reduce the particle bounce. It is important to characterize the CV performance in different environments. In this study, we characterized specific organic aerosols (OAs) from vehicle, cooking, biomass burning, and coal burning emissions by a time-of-flight ACSM (TOF-ACSM) with the CV. Their corresponding marker ions that have been defined in the previous SV-based analysis are still valid in the CV mass spectra. Spectra of OAs from cooking and vehicle exhaust show similarities in distinct alkyl fragments but different ratios of m∕z 55 and 57. Ions related to polycyclic aromatic hydrocarbons are present in the OA spectra obtained from burning lignite and bituminous coal but not in the spectra obtained from burning anthracite. Although the relative intensities of m∕z 60 and 73 are much lower in the CV spectra than in the SV spectra for biomass burning OA, they are still relatively greater compared with the spectra for other sources. Our data suggest an atmospheric background of f60 of below 0.03 % for CV. Moreover, we deployed the CV TOF-ACSM along with a SV AMS in urban Beijing during the winter of 2017 to characterize ambient OA with strong anthropogenic influences. The CV TOF-ACSM shows a collection efficiency (CE) of about unity. The CV and SV data show consistent mass concentrations of sulfate, nitrate, ammonium, and OA. Six OA factors are identified by the positive matrix factorization (PMF) analysis for both the CV and the SV data. The SV and CV PMF factors show good correlations in mass concentrations. The SV and CV factors related to coal combustion and cooking differ significantly in loadings, explained by the PMF uncertainty and the lack of understanding of the relative ionization efficiency and CE for primary OA. The CV factors related to secondary sources show greater loadings than the SV factors, which may be associated with the changes in signal-to-noise ratios of various ions in the PMF analysis. Our results support improved mass quantification and useful source identification by the CV for ambient particles in the polluted urban environment. The difference in factor loadings between SV and CV should be considered when interpreting or comparing the PMF results among studies.
12 October 2023 in Environmental Science & Technology
One strategy for mitigating the indoor transmission of airborne pathogens, including the SARS-CoV-2 virus, is irradiation by germicidal UV light (GUV). A particularly promising approach is 222 nm light from KrCl excimer lamps (GUV222); this inactivates airborne pathogens and is thought to be relatively safe for human skin and eye exposure. However, the impact of GUV222 on the composition of indoor air has received little experimental study. Here, we conduct laboratory experiments in a 150 L Teflon chamber to examine the formation of secondary species by GUV222. We show that GUV222 generates ozone (O3) and hydroxyl radicals (OH), both of which can react with volatile organic compounds to form oxidized volatile organic compounds and secondary organic aerosol particles. Results are consistent with a box model based on the known photochemistry. We use this model to simulate GUV222 irradiation under more realistic indoor air scenarios and demonstrate that under some conditions, GUV222 irradiation can lead to levels of O3, OH, and secondary organic products that are substantially elevated relative to normal indoor conditions. The results suggest that GUV222 should be used at low intensities and in concert with ventilation, decreasing levels of airborne pathogens while mitigating the formation of air pollutants.
Automatic radiology report generation has attracted enormous research interest due to its practical value in reducing the workload of radiologists. However, simultaneously establishing global correspondences between the image (e.g., Chest X-ray) and its related report and local alignments between image patches and keywords remains challenging. To this end, we propose an Unify, Align and then Refine (UAR) approach to learn multi-level cross-modal alignments and introduce three novel modules: Latent Space Unifier (LSU), Cross-modal Representation Aligner (CRA) and Text-to-Image Refiner (TIR). Specifically, LSU unifies multimodal data into discrete tokens, making it flexible to learn common knowledge among modalities with a shared network. The modality-agnostic CRA learns discriminative features via a set of orthonormal basis and a dual-gate mechanism first and then globally aligns visual and textual representations under a triplet contrastive loss. TIR boosts token-level local alignment via calibrating text-to-image attention with a learnable mask. Additionally, we design a two-stage training procedure to make UAR gradually grasp cross-modal alignments at different levels, which imitates radiologists’ workflow: writing sentence by sentence first and then checking word by word. Extensive experiments and analyses on IU-Xray and MIMIC-CXR benchmark datasets demonstrate the superiority of our UAR against varied state-of-the-art methods.
20 December 2022 in Atmospheric Chemistry and Physics
The atmospheric oxidation of dimethyl sulfide (DMS) represents a major natural source of atmospheric sulfate aerosols. However, there remain large uncertainties in our understanding of the underlying chemistry that governs the product distribution and sulfate yield from DMS oxidation. Here, chamber experiments were conducted to simulate gas-phase OH-initiated oxidation of DMS under a range of reaction conditions. Most importantly, the bimolecular lifetime (τbi) of the peroxy radical CH3SCH2OO was varied over several orders of magnitude, enabling the examination of the role of peroxy radical isomerization reactions on product formation. An array of analytical instruments was used to measure nearly all sulfur-containing species in the reaction mixture, and results were compared with a near-explicit chemical mechanism. When relative humidity was low, “sulfur closure” was achieved under both high-NO (τbi<0.1 s) and low-NO (τbi>10 s) conditions, though product distributions were substantially different in the two cases. Under high-NO conditions, approximately half the product sulfur was in the particle phase, as methane sulfonic acid (MSA) and sulfate, with most of the remainder as SO2 (which in the atmosphere would eventually oxidize to sulfate or be lost to deposition). Under low-NO conditions, hydroperoxymethyl thioformate (HPMTF, HOOCH2SCHO), formed from CH3SCH2OO isomerization, dominates the sulfur budget over the course of the experiment, suppressing or delaying the formation of SO2 and particulate matter. The isomerization rate constant of CH3SCH2OO at 295 K is found to be 0.13±0.03 s−1, in broad agreement with other recent laboratory measurements. The rate constants for the OH oxidation of key first-generation oxidation products (HPMTF and methyl thioformate, MTF) were also determined ( cm3 molec.−1 s−1, cm3 molec.−1 s−1). Product measurements agree reasonably well with mechanistic predictions in terms of total sulfur distribution and concentrations of most individual species, though the mechanism overpredicts sulfate and underpredicts MSA under high-NO conditions. Lastly, results from high-relative-humidity conditions suggest efficient heterogenous loss of at least some gas-phase products.
The recent video grounding works attempt to introduce vanilla contrastive learning into video grounding. However, we claim that this naive solution is suboptimal. Contrastive learning requires two key properties: (1) alignment of features of similar samples, and (2) uniformity of the induced distribution of the normalized features on the hypersphere. Due to two annoying issues in video grounding: (1) the coexistence of some visual entities in both ground truth and other moments, i.e. semantic overlapping; (2) only a few moments in the video are annotated, i.e. sparse annotation dilemma, vanilla contrastive learning is unable to model the correlations between temporally distant moments and learned inconsistent video representations. Both characteristics lead to vanilla contrastive learning being unsuitable for video grounding. In this paper, we introduce Geodesic and Game Localization (G2L), a semantically aligned and uniform video grounding framework via geodesic and game theory. We quantify the correlations among moments leveraging the geodesic distance that guides the model to learn the correct cross-modal representations. Furthermore, from the novel perspective of game theory, we propose semantic Shapley interaction based on geodesic distance sampling to learn fine-grained semantic alignment in similar moments. Experiments on three benchmarks demonstrate the effectiveness of our method.