Enhancing low-light images with machine learning-based techniques becomes increasingly difficult as resolution rises. To address this challenge, researchers from China have developed LL-Refiner, a novel framework that first performs coarse enhancement at low resolution before progressively refining details at multiple scales. Tests on real-world datasets show that LL-Refiner achieves state-of-the-art image quality while also improving performance in computer-vision tasks such as depth estimation. 

Why do some tissues regenerate efficiently after injury while others develop chronic inflammation, scar formation, or even fibrosis? A newly published review proposes that immunometabolism is not a universal "pro-inflammatory or anti-inflammatory" program but a dynamic adaptation process shaped by organ structure, regenerative capacity, injury stage, and local microenvironment. The article systematically compares six tissues—intestine, skin, liver, lung, heart, and skeletal muscle—and introduces the concept of niche-dependent immunometabolic regulation.

More than two-thirds of hospitalized COVID-19 survivors maintained stable cognition 2 years after Omicron infection, while one in five experienced reversible decline. Researchers report that only 10.8% developed irreversible cognitive decline, including 3.6% with progressive decline. Older age, reinfection, severe illness, hypertension, longer hospitalization, and white matter lesions increased the risk, while education and vaccination were protective. Findings highlight potential for cognitive recovery, supporting monitoring of patients at higher risk of progressive decline.

By regulating coordination angles of two carbazole-pyridine ligands (L1, L2) and building blocks (E1(OTf)₂, E2(OTf)₂), successfully obtained four architectures: a molecular-tweezer-like compound, a snowflake-shaped fold assembly, an 8₁₈ knot, and a closed three-link chain. SC-XRD validated the structures and showed that the ligand angle determines whether the assembly is interlocked or folded. NMR and ESI-TOF-MS verified solution-phase formation and purity. Furthermore, the closed three-link chain exhibits high photothermal conversion performance. Its efficiency is directly influenced by π–π stacking interactions, and EPR experiments further confirm that the radical character of the building blocks also contributes to the photothermal properties.

A research team led by Professor Kaicun Wang at Peking University has released a homogenized gridded climate dataset for China (HCD01), providing daily, 0.1-degree climate data for 1961-2022. The dataset corrects non-climatic signals in long-term observations and merges homogenized station records with ERA5-Land reanalysis, offering a high-quality data basis for studying climate change and its impacts across China.

Researchers have developed a membrane-based strategy that separates complex light naphtha into valuable chemical streams at the molecular level. By mildly etching a metal–organic framework membrane with tannic acid, the team created complementary membranes that sort hydrocarbons by size and chemical affinity. The membrane cascade separated a 15-component petroleum mixture into three product streams while potentially reducing energy consumption by about 91% compared with conventional distillation.

In a paper published in SCIENCE CHINA Earth Sciences, researchers assessed the parameterization of orographic gravity wave drag (OGWD) in 3-km-resolution WRF simulations for precipitation forecast over the highly complex-terrain Southwest China. Their results reveal that OGWD parameterization, previously considered negligible in kilometer-scale models, can greatly improve the predicted rainfall intensity and location by modulating the low-level winds and moisture transport. These findings lay an important scientific foundation for operational precipitation forecasting in regions of complex terrain.

The integration of luminescent nanomaterials into scalable semiconductor platforms is vital for on-chip photonics. This study demonstrates an approach to fabricate fluorescent nanostructures by hybridising carbon dots (CDs) with semiconductor nanowires (NWs) grown on Si. By coupling the tuneable chemistry of CDs with the tailored optoelectronics of semiconductor NWs, this study establishes a scalable host–guest architecture for engineering nanoscale light sources to promote applications in integrated photonic circuits and sensing.

Energy-efficient visible-band intelligent vision processors demands both optical neural network (ONN) design and high-speed nanofabrication advances. Chinese researchers proposed a random-projection ONN design strategy and established a multi-focus random-access nanofabrication platform that fabricates visible-light ONN devices with 4 million optical neurons per mm² in just 15 minutes, hitting a new record. Miniaturized camera integration ensures high-accuracy complex vision tasks with low power consumption, advancing optical sensing, medical diagnosis and human-computer interaction.