STEM与日常科技·英语精读30篇（5）
9 / 30
正在确认阅读权限…
LiDAR Point Cloud Degradation in Rain and Fog: Physics-Based Modeling of Signal Attenuation
激光雷达点云在雨雾中的退化：基于物理的信号衰减建模
-
Raindrops scatter 905 nm laser pulses through Mie scattering, reducing point cloud density by up to 73% at 25 mm/h intensity—degradation that scales nonlinearly with droplet size distribution, not just rainfall rate.雨滴通过米氏散射使905 nm激光脉冲发生散射，降雨强度达25 mm/h时点云密度最多下降73%——性能退化程度与雨滴尺寸分布呈非线性关系，而不仅取决于降雨率。
-
Fog attenuation follows exponential decay governed by Beer-Lambert law, but standard models underestimate backscatter from polydisperse aerosols, leading to false ‘free-space’ classifications in autonomous vehicle perception stacks.雾引起的衰减遵循比尔-朗伯定律的指数衰减规律，但标准模型低估了多分散气溶胶的后向散射，导致自动驾驶感知系统误判为‘自由空间’。
-
Wavelength choice matters critically: 1550 nm lasers penetrate fog better than 905 nm but require more expensive InGaAs detectors and face stricter Class 1 eye-safety limits, constraining maximum pulse energy.波长选择至关重要：1550 nm激光比905 nm穿透雾的能力更强，但需更昂贵的InGaAs探测器，且受更严格的Class 1人眼安全限值约束，限制了最大脉冲能量。
-
Real-time compensation requires dual-wavelength systems—measuring extinction at both bands to estimate Mie-to-Rayleigh scattering ratios—but adds optical complexity and calibration drift over thermal cycles.实时补偿需采用双波长系统——同步测量两波段消光以估算米氏/瑞利散射比——但增加了光学复杂度，并在热循环中易出现标定漂移。
-
Point cloud denoising algorithms trained on clear-weather data fail catastrophically in precipitation, misclassifying rain-induced speckle as static obstacles or occluding true pedestrians behind wet windshield streaks.在晴天数据上训练的点云去噪算法在降水场景中彻底失效，将雨滴引起的散斑误判为静态障碍物，或因挡风玻璃水痕遮挡而漏检真实行人。
-
Sensor fusion mitigates but doesn’t eliminate risk: camera-based depth estimation degrades in low contrast, while radar lacks angular resolution to distinguish overlapping vehicles in dense traffic.传感器融合可缓解但无法消除风险：基于相机的深度估计在低对比度下性能下降，而雷达角分辨率不足，难以在密集车流中区分重叠车辆。
-
Automotive safety standards like ISO 21448 (SOTIF) now mandate rain/fog validation using calibrated nozzles that replicate droplet spectra—not just uniform water films.ISO 21448（SOTIF）等汽车安全标准现已强制要求使用校准喷嘴模拟真实雨滴谱（而非均匀水膜）开展雨雾验证。
-
Field data shows that lidar range reliability drops from 200 m in dry air to under 45 m in moderate fog (visibility 50 m), forcing fallback to conservative longitudinal control policies.实测数据显示，激光雷达在干燥空气中的有效探测距离为200 m，而在中等雾况（能见度50 m）下骤降至不足45 m，迫使系统启用保守的纵向控制策略。
-
Physics-informed neural networks now embed Mie scattering cross-sections as hard constraints, improving generalization—but require GPU-accelerated ray tracing during inference.物理信息神经网络现将米氏散射截面作为硬约束嵌入模型，提升了泛化能力——但推理阶段需GPU加速的光线追踪。
-
Ultimately, robust autonomy demands not just better sensors but redefined operational design domains—acknowledging that some weather conditions remain fundamentally unperceivable with current photon budgets.最终，实现鲁棒自动驾驶不仅依赖更优传感器，更需重新定义运行设计域——承认当前光子预算下，某些天气条件本质上不可感知。