Search

Nitrogen (N) doping is an effective strategy for improving the solar-driven photocatalytic performance of anatase TiO2, but controllable methods for nitrogen-rich doping and associated defect engineering are highly desired. In this work, N-rich doped anatase TiO2 nanoparticles (4.2 at%) were successfully prepared via high-temperature nitridation based on thermally stable H3PO4-modified TiO2. Subsequently, the associated deep-energy-level defects such as oxygen vacancies and Ti3+ were successfully healed by smart photo-Fenton oxidation treatment. Under visible-light irradiation, the healed N-doped TiO2 exhibited a ~2-times higher activity of gas-phase acetaldehyde degradation than the non-treated one and even better than standard P25 TiO2 under UV-visible-light irradiation. The exceptional performance is attributed to the extended spectral response range from N-rich doping, the enhanced charge separation from hole capturing by N-doped species, and the healed defect levels with the proper thermodynamic ability for facilitating O2 reduction, depending on the results of ∙O2− radicals and defect measurement by electron spin resonance, X-ray photoelectron spectroscopy, atmosphere-controlled surface photovoltage spectra, etc. This work provides an easy and efficient strategy for the preparation of high-performance solar-driven TiO2 photocatalysts.

Convolutional Neural Networks (CNNs) have emerged as a leading model in hyperspectral image (HSI) classification, owing to their unique features such as parameter sharing and localized receptive fields. Despite these advantages, current CNN-based approaches face significant challenges, including complex model architectures, the inability to fully extract deep features and the underutilization of CNNs kernels. This study introduces a novel multidimensional sequential CNN model, termed the 3D-1D-2D SCNN, specifically designed for HSI classification. The model begins with Principal Component Analysis (PCA) to reduce the dimensionality of the HSI data. It then sequentially applies multidimensional convolutions to extract a diverse array of features from the HSI data. The classification process is finalized using a linear classifier, ensuring precise categorization of the hyperspectral data. Simulation results demonstrate that this model outperforms traditional methods in classification accuracy across four widely-used HSI datasets. [ABSTRACT FROM AUTHOR]