Scene text image super-resolution （STISR） aims to recover clear and recognizable high-resolution text content from blurry and degraded low-resolution scene text images， serving as an important research topic that bridges low-level visual restoration and high-level text understanding. This paper presents a systematic review of recent advances in STISR. First， we review the development of general image super-resolution methods and analyze their limitations when directly applied to low-resolution scene text images. Second， we focus on extensively studied English scene text image super-resolution methods and analyze the main characteristics and evolution of different technical paradigms. On this basis， we further summarize recent progress in multilingual scene text image super-resolution and text-aware image super-resolution for full natural scenes. Subsequently， commonly used datasets and evaluation metrics in this field are reviewed， and the performance of representative methods is systematically compared. Finally， we summarize the current challenges from the perspectives of visual quality and text recognition evaluation， generalization， hallucination， model adaptation， and practical deployment， and discuss promising directions for future research.
To improve the predictive accuracy of Reynolds-averaged Navier-Stokes （RANS） models for complex flows such as separation and secondary flows， a data-driven approach is used to correct RANS predictions. Based on the tensor representation theory， a constitutive relationship between the nonlinear Reynolds force vector （RFV） and a vector basis is established. High-fidelity flow data and a frozen-flow-field method are then employed to extract input features and the nonlinear RFV. A nonlinear RFV prediction model is constructed using a vector-basis neural network and subsequently coupled with the flow solver. Results demonstrate that the proposed model enhances the shear stress transport （SST） model’s ability to capture secondary flows in square duct flow and improves its prediction of flow separation in periodic hill flow. The coupled framework achieves computational efficiency comparable to conventional RANS simulations while significantly improving accuracy， and it exhibits good convergence as well as strong generalization with respect to geometry and Reynolds number.
Modeling real‑world airside operational data is difficult due to significant noise interference， diverse operating mechanisms， and the challenge of explicitly defining key semantic boundaries. These issues often cause attention drift， unstable prediction errors， and insufficient representation of operational semantics. To address these issues， this study proposes a fuzzy domain knowledge-enhanced multi‑branch Transformer （DK-MBT） for predicting airside operational conditions. Expert knowledge is embedded into feature representations using fuzzy sets， with membership functions that encode the interval semantics of critical indicators. These functions are further equipped with learnable mechanisms for adaptive adjustment that introduce semantic constraints into the input sequence. Additionally， a fuzzy-semantics-modulated multi-branch attention structure is designed to correct attention scores via a fuzzy relation matrix and suppress noise-induced attention drift. Moreover， parallel branches capture various temporal interaction patterns， and their fused representations enhance the complex operational modeling. Experimental results demonstrate that DK-MBT outperforms existing methods， achieving more robust predictive performance on two regression tasks， namely taxi-out time and departure delay. Moreover， feature attribution based on SHapley additive explanations （SHAP） is applied to post-hoc analysis of model outputs， in order to observe the contribution of different operational factors to the prediction results.
Extreme overexposure causes luminance saturation， detail loss， and degradation of chroma cues. Existing exposure enhancement methods can restore visibility and luminance structures to some extent， but most of them jointly adjust luminance and color in the sRGB space. As a result， residual color casts may still appear in saturated or low-chroma regions， leading to global tone shifts and local color distortions. To address this problem， a luminance-preserving chroma correction network （LCC-Net） is proposed for color casts in extreme overexposure results. The original input image and its luminance-adjusted result are mapped into the CIELAB space. LCC-Net preserves the corrected L channel and predicts only the a、b chroma components， avoiding disturbance to the recovered luminance structure. An overexposure area mask， a global histogram-based chroma encoder， and a local attentive chroma correction module are further introduced to locate unreliable chroma regions， constrain the global tone distribution， and correct spatially varying local color casts. Experimental results show that LCC-Net works with various frozen luminance adjusters and improves color naturalness and consistency in extreme overexposure scenarios.
Vertical trajectory optimization is an important technical approach for improving airline operational efficiency， reducing fuel consumption and promoting green aviation. To address the challenges of strong nonlinearity， multi-parameter and high computational complexity， a novel vertical trajectory optimization method based on a nonlinear cosine adaptive perturbation genetic algorithm is proposed， aiming to minimize the direct operating cost. An optimal cost model is established using the energy state method， and key trajectory parameters are optimized through nonlinear cosine adaptive perturbation genetic algorithm （NCAP-GA）. Nonlinear cosine-based adaptive crossover and mutation strategies， together with a stagnation-triggered perturbation mechanism， are designed to enhance population diversity， global search capability and convergence efficiency. Comparative results demonstrate that the proposed algorithm outperforms genetic algorithm （GA）， adaptive genetic algorithm （AGA）， and partical swarm optimization （PSO） in terms of optimization accuracy and convergence speed. Simulation results further verify significant reductions in fuel consumption and flight time， confirming the effectiveness and engineering applicability of the proposed method.
