arXiv subjects

Shuoxun Zhang

Publications and source records attributed to Shuoxun Zhang.

4 recordsLinked to original sources

arXiv2026

Advancing MLLM-based UAV Image Understanding and Reasoning: A Benchmark and a Training-Free Multi-Agent System

Multimodal Large Language Model (MLLM)-based UAV aerial image understanding and reasoning is essential for aerial intelligence yet poses distinct challenges arising from extreme scale variation, arbitrary camera orientations, and high object density. Despite growing interest, existing evaluations remain fragmented across individual datasets and narrow tasks, leaving a critical gap in unified assessment of UAV understanding and reasoning capabilities. To fill this gap, we construct UAVQA-Bench, a benchmark of 1,500 human-annotated QA pairs drawn from 13 public UAV datasets, covering 6 capability dimensions and 16 tasks in both multiple-choice and visual grounding formats. Systematic evaluation of a broad range of open-source and closed-source MLLMs as well as agent-based systems on UAVQA-Bench identifies three key failure modes: domain-toolset mismatch, unchecked error propagation, and static reasoning. Motivated by these findings, we propose UAV-MAS, a training-free multi-agent system for MLLM-based UAV aerial image understanding and reasoning, comprising a Domain-Specific Perception Engine (DSPE) that routes queries to task-appropriate visual tools, a Context-Aware Iterative Refinement module (CAIR) that validates intermediate reasoning to curb error accumulation, and a Difficulty-Aware Adaptive Search mechanism (DAAS) that adjusts search depth to question difficulty. UAV-MAS with a 32B open-source MLLM achieves 77.0% overall accuracy on UAVQA-Bench, surpassing Gemini 3 Pro by 4.0\%, while the 8B variant improves 8.7\% over its base model.

cs.CV↗

arXiv2026

Pulsation of Burner-Stabilized CH4-O2 Flames Moderated by CO2 Addition

This study investigated the pulsating instability of burner-stabilized premixed CH4-O2 flames at various levels of CO2 dilution. Experiments were conducted using a water-cooled porous-plug burner of 18 mm diameter over a wide range of mixture compositions and flow rates, during which time-resolved measurements of flame chemiluminescence and gas temperature were obtained. The primary oscillation frequencies of the pulsating flames were determined using fast Fourier transform and harmonic power analysis. Phase-locked analysis of the chemiluminescence images revealed an interesting mode-transition phenomenon of the flame oscillations. Under fuel-rich conditions with relatively low heat release rates and low flow rates, the flames exhibited quasi-periodic single-mode oscillations. At elevated flow rates, these oscillations were modulated by low-frequency flame flickering instabilities, which created sidebands around the primary oscillation frequency. At higher heat release rates, the flickering instability further triggered mode splitting, eventually leading to multi-mode oscillations. Regime diagrams of the flame oscillation modes, as well as the stability boundaries, were obtained under various fuel flow rates. These findings can be useful for both fundamental research on flame dynamics and practical applications of CO2-moderated oxy-combustion.

physics.flu-dyn↗

arXiv2025

Precision Thermometry of Flat Flames Using Spatially Resolved Multi-Color Laser Absorption Spectroscopy of Carbon Dioxide

This work developed an accurate and robust absorption-based method for spatially resolved measurements of gas temperatures in flames and reacting flows, with typical single-measurement uncertainties on the order of 1\%. This method exploits narrow-linewidth laser absorption of hot CO$_2$ molecules, which can be generated from combustion or artificially seeded into the flow. A collinear dual-laser setup allowed for periodic scans over tens of CO$_2$ absorption transitions near the $ν_3$ bandhead every 100 $μs$, from which gas temperatures (as well as CO$_2$ concentrations) were determined with high sensitivity and robustness. Spatially resolved measurements were achieved using an electrically driven high-speed beam scanning system consisting of a 2-D galvo scanner and a pair of off-axis parabolic mirrors. An effective spatial resolution of 1 mm was achieved at a planar field measurement speed of 200 Hz and a volumetric field measurement speed of 2 Hz. A physically constrained nonlinear inference framework was also developed for the quantitative analysis of the measurement data. Proof-of-concept experiments were performed on axisymmetric flames stabilized on a Mckenna burner at various equivalence ratios and flow rates, and the results agreed asymptotically with the theoretical value of the adiabatic flame temperature. An additional experiment on a flame of complex geometry demonstrated an excellent level of resolution, precision, and contrast achieved by the current thermometry method. This method promises to provide good utility in future combustion studies due to its high performance metrics and relative ease of use.

physics.optics↗

arXiv2025

Quantitative 3D Morphology of Cellular H2/O2/N2 Flames on a Porous-Plug Burner: Spatially Resolved Measurements of Temperature and OH Radical

This study presents a systematic characterization of burner-stabilized lean hydrogen flame morphology across a wide range of equivalence ratios, dilution factors, and flow rates. Spatially resolved measurements of three-dimensional temperature and OH distributions were achieved. A comprehensive dataset of over 200 flame cases was obtained, enabling accurate determination of regime diagrams for different flame modes. Linear stability analysis and direct numerical simulations were also performed and compared with the experimental results. The dominant wavenumbers of steady-state cellular flames were found to be consistently lower than the most unstable wavenumbers predicted by the linearized dispersion relation, indicating that nonlinear interactions between finite-amplitude perturbations of different length scales favored the growth of low-frequency components at long times. The cellular structures were found to be critically important in stabilizing the flame, especially at nominal equivalence ratios near the lean flammability limit. The mechanism of cellular flame stabilization was analyzed by complementary numerical simulations using a detailed reaction model. The combined effect of curvature-induced flame acceleration, local flow expansion/compression near the burner surface, and stratification of equivalence ratio caused by Soret diffusion created regions of reduced flow speed and enriched hydrogen concentration that helped anchor flames at nominal conditions where they would have blown off without the flame cells. The results of the present study are useful for understanding the fundamental flame dynamics of lean hydrogen mixtures and for improving the design of practical hydrogen combustors.

physics.flu-dyn↗