数据科学 / 机器学习
Data Science / Machine Learning
Data Science / Machine Learning
我们急需一名数据科学家，有空间科技经验者优先，但非必需。# 数据科学家，最好有地理空间/遥感专业背景
## 关于职位 我们正在构建一个基于卫星的早期预警系统，利用Sentinel-1（SAR）和Sentinel-2（光学）影像、基础模型和物理沙丘迁移建模，检测风驱动的沙尘侵袭和沙漠管道网络中的管道位移。你将负责检测和校准科学：将原始卫星数据转化为经过校准、验证的警报，并与实际现场记录事件进行对比。
你将负责的工作 - 在管道走廊的多时相Sentinel-1/2影像上构建和校准变化检测和异常模型。 - 学习各站点的“正常地形”基线，并根据地面真实事件日志验证检测结果（检测率、提前时间、误报率、AUC）。 - 在有限标注数据上使用LoRA/PEFT对地理空间基础模型（Prithvi-EO / 类似模型）进行微调。 - 实现SAR技术用于位移检测：幅度变化、相干性以及像素偏移跟踪，以测量管道和沙丘的移动。 - 开发沙丘迁移追踪（光流/特征追踪）、迁移方向和移动性指标。 - 工程化从Copernicus（CDSE / Sentinel Hub / STAC）获取数据，并融合光学、SAR、DEM和ERA5风数据。 - 设计标注策略（侵袭掩码、严重程度）和训练/验证划分，避免数据泄露。 - 向技术和业务利益相关者诚实传达结果和限制。
要求 - 7年以上应用数据科学/机器学习经验，具备实际地理空间遥感经验。 - 熟练使用Python：numpy、rasterio/GDAL、xarray、scikit-image、geopandas/shapely。 - 理解光学和SAR数据（光谱指数、后向散射/dB、分辨率权衡、重访周期）。 - 深度学习使用PyTorch；有微调模型的经验（迁移学习、LoRA/PEFT）。 - 模型验证和校准：ROC/AUC、阈值设定、交叉验证、处理弱标签或少量标签。 - 时间序列/变化检测方法和坐标参考系统（UTM、重新投影）。
加分项 - InSAR/SAR偏移跟踪（SNAP、ISCE或类似工具）用于地表/结构位移。 - 地理空间基础模型（Prithvi-EO、TerraTorch、HLS）和分割。 - Copernicus/CDSE、Sentinel Hub、STAC、Planetary Computer。 - 风蚀地貌/沙丘动力学；石油和天然气或管道行业经验。
查看英文原文
Data Science / Machine Learning
We have an urgent requirement for a Data Scientist, preferably with Space Tech experience but not mandatory.# Data Scientist preferably with specialization in Geospatial / Remote Sensing
## About the role We are building a satellite-based early-warning system that detects wind-driven sand encroachment and pipeline displacement across a desert pipeline network, using Sentinel-1 (SAR) and Sentinel-2 (optical) imagery, foundation models, and physical dune-migration modelling. You will own the detection and calibration science: turning raw satellite passes into calibrated, validated alerts against real field-logged events.
What you will do- Build and calibrate change-detection and anomaly models on multi-temporal Sentinel-1/2 imagery over pipeline corridors. - Learn per-site """"normal terrain"""" baselines and validate detections against a ground-truth event log (detection rate, lead time, false-positive rate, AUC). - Fine-tune geospatial foundation models (Prithvi-EO / similar) with LoRA/PEFT on limited labelled data. - Implement SAR techniques for displacement: amplitude change, coherence, and pixel-offset tracking to measure pipe and dune movement. - Develop dune-migration tracking (optical flow / feature tracking), migration direction, and mobility indices. - Engineer robust ingestion from Copernicus (CDSE / Sentinel Hub / STAC) and fuse optical, SAR, DEM, and ERA5 wind data. - Design labelling strategy (encroachment masks, severity) and a train/validation split that avoids leakage. - Communicate results and limitations honestly to technical and business stakeholders.
Required -7+ years applied data science / ML, with hands-on geospatial remote sensing. - Strong Python: numpy, rasterio/GDAL, xarray, scikit-image, geopandas/shapely. - Working knowledge of optical and SAR data (spectral indices, backscatter/dB, resolution trade-offs, revisit). - Deep learning with PyTorch; experience fine-tuning models (transfer learning, LoRA/PEFT). - Model validation and calibration: ROC/AUC, thresholding, cross-validation, handling weak/few labels.- Time-series / change-detection methods and coordinate reference systems (UTM, reprojection).
Nice to have - InSAR / SAR offset tracking (SNAP, ISCE, or equivalent) for surface/structure displacement. - Geospatial foundation models (Prithvi-EO, TerraTorch, HLS) and segmentation. - Copernicus/CDSE, Sentinel Hub, STAC, Planetary Computer. - Aeolian geomorphology / dune dynamics; oil & gas or pipeline-integrity domain exposure. - MLOps and cloud (containerisation, scheduled inference, geospatial data pipelines).
Qualifications - MSc/PhD in Remote Sensing, Geospatial Science, Earth Observation, CS/ML, Physics, or equivalent experience
ADVANCED ENGLISH
REMOTE
Originally posted on Himalayas