- Research Article
- 10.1016/j.patcog.2026.113449
PointSlice: Accurate and efficient slice-based representation for 3D object detection from point clouds
- Oct 01, 2026
- Pattern Recognition
- Qifeng Liu + 9 more +9
Publications from 2021 to 2026
Showing 10 of 40,296 papers
PointSlice: Accurate and efficient slice-based representation for 3D object detection from point clouds
Normalized solutions for Kirchhoff-type problems with critical exponential growth in bounded domains
Manifold microchannel heat sinks enhance the power generation performance of micro-thermoelectric devices
• Ultrahigh power factor boosts maximum power under a fixed temperature difference (Δ T ). • Manifold microchannel (MMC) heat sink enables a large Δ T in power generation mode. • Mg 3 Sb 2 and SnSe are potential alternatives to Bi 2 Te 3 . • Micro devices with MMC heat sinks offer promising energy harvesting. Micro thermoelectric devices (micro-TEDs) have emerged as promising candidates for energy harvesting applications. Despite the development of numerous high-performance thermoelectric materials over the past decades, most micro-TEDs still rely heavily on conventional Bi 2 Te 3 -based materials. Effectively integrating advanced thermoelectric materials into micro-TEDs remains a significant challenge. In this work, we theoretically match several recently developed high-performance near-room-temperature thermoelectric materials with micro-TEDs and explore the integration of a manifold microchannel (MMC) heat sink on the cold side to significantly enhance the effective temperature difference across the devices. Upon incorporating the MMC heat sink, the maximum power densities of Bi 2 Te 3 -, Mg 3 Sb 2 -, and SnSe-based micro-TEDs reach 50.32, 29.68, and 31.07 µW mm −2 , respectively, representing increases of 60.63-fold, 56.52-fold, and 57.69-fold compared to devices without heat sinks. These results highlight that coupling micro-TEDs with emerging near-room-temperature thermoelectric materials and a well-engineered MMC heat sink offers a promising route for efficient energy harvesting applications.
Read moreSynthesis of Tb3Al5O12 nano powders via nitrate decomposition and self-seeding
An Integrated Visible-Light Communication/Gesture-Recognition System Enabled by Machine Learning
Integrating high-speed visible light communication (VLC) with natural human–machine interaction is important for smart lighting and secure indoor Internet of Things (IoT) systems, yet most gesture interfaces rely on cameras or extra sensors, raising privacy concerns and system complexity. This paper presents a waveform-sharing, dual-task VLC receiver that jointly demodulates data and recognizes gestures from the same photoelectric waveform without additional sensing hardware, enabling camera-free, privacy-preserving interactive VLC. Machine learning (ML) is employed for end-to-end compensation and classification: a convolutional neural network (CNN)-based post-equalizer mitigates time-varying distortions from lightemitting diode (LED) nonlinearity, multipath propagation, and gesture occlusion, boosting the raw rate of a 1 m indoor blue-LED link from 30 Mb/s to 46 Mb/s in the communication-only setting, while spectro-temporal features with a lightweight classifier achieve ~98% accuracy over eight dynamic gestures. With gesture inference enabled, the net payload rate remains ~26 Mb/s with negligible added latency, demonstrating that interaction can be overlaid with limited throughput loss and reduced power, cost, and privacy risk.
Read moreEffect of stress state and strain rate on macro–micro damage in Ti-6Al-4V alloy: Experiments and theoretical model
A flux solver based on discrete Boltzmann method and its application to compressible flows
Service reliability of intelligent computing cluster systems for large language models
Multi-action predictive maintenance for lithium-ion batteries based on stacked bilinear neural network
Active Shimming Magnet for Dipole Accelerator to Compensate Sextupole Harmonic Field
An active shimming magnet was developed to compensate sextupole error fields in a block-type REBCO dipole magnet. The magnet employs a sextupole configuration with three no-insulation double-pancake coils wound from 2 mm-wide REBCO tapes. Standalone tests at 77 K confirmed intact electromagnetic performance after winding and assembly. Field quality measurements at 4.2 K under dipole fields of 1.0 T and 3.1 T showed a linear decrease of sextupole component <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$b_{3}$</tex-math></inline-formula> with increasing shimming current up to 30 A, with consistent compensation ability at both fields. The active shimming magnet caused an increase in <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$b_{5}$</tex-math></inline-formula> and a decrease in <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$b_{7}$</tex-math></inline-formula>, both with magnitudes much smaller than that of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$b_{3}$</tex-math></inline-formula>, indicating limited impact on overall field quality. These results demonstrate that the active shimming magnet effectively reduces sextupole field errors to within 1 unit, providing a compensation approach for improving field quality in future HTS accelerator dipole magnets.
Read more