About Us
SkyServe is creating the intelligence layer for orbital and terrestrial observation. In our Maritime Domain Awareness (MDA) vertical, tracking non-cooperative and illicit maritime actors requires looking beyond broadcast AIS signals. A vast percentage of illegal, unreported, and unregulated (IUU) fishing, maritime smuggling, and sanctions evasions occur when vessels deliberately switch off their AIS transponders ("go dark").To solve this, our computer vision pipeline ingests and processes high-volume satellite imagery—including Synthetic Aperture Radar (SAR) (Sentinel-1, ICEYE, Capella, Umbra) and Electro-Optical (EO) / Multispectral imagery (Sentinel-2, PlanetScope, Maxar, commercial optical constellations)—to autonomously detect, classify, and track vessels across global sea lanes day or night, regardless of cloud cover.
Position Overview
As a Senior Computer Vision Engineer (EO, SAR, etc.), you will design, benchmark, and deploy deep learning vision models that detect small maritime targets, segment vessel wakes to estimate speed and heading, classify vessel types, and correlate visual satellite detections against broadcast RF/AIS data to instantly expose "dark" ships.
Key Responsibilities
1. SAR & EO Vision Model Development
Small-Target Ship Detection: Develop high-precision object detection models (oriented bounding boxes, rotated YOLO, Faster R-CNN, Vision Transformers) specifically optimized for tiny, low-pixel vessel signatures across large ocean swaths.SAR Feature Extraction & Speckle Handling: Design neural architectures capable of handling complex SAR speckle noise, varying sea-state clutter, polarization channels (VV, VH, HH, HV), and incident angles to identify metallic radar cross-sections (RCS).Wake Detection & Kinematics Estimation: Implement segmentation models to isolate Kelvin ship wakes and turbulent water trails, extracting vessel heading, velocity, and length directly from imagery.Fine-Grained Classification: Build multi-class classifiers to categorize detected hulls into merchant vessels, container ships, tankers, tugs, naval assets, and artisanal/industrial fishing craft.2. Multi-Sensor Cross-Modal Fusion
Work with data engineering teams to coregister and fuse optical EO, SAR, and RF geolocations with spatio-temporal AIS tracks.Build association logic to perform Dark Vessel Matching—identifying satellite-detected ship coordinates that lack an active AIS broadcast within a defined spatial and temporal threshold.Leverage cloud masking, sea-land segmentation, and bathymetric data to eliminate false positives (islets, oil platforms, wind farms, cloud edges).3. Large-Scale Geospatial Pipeline Engineering
Design efficient tiling and inference workflows to process giga-pixel satellite swaths without memory overflows or truncation artifacts.Curate, augment, and manage multi-sensor maritime datasets using active learning, synthetic SAR/EO simulations, and automated pseudo-labeling.Enforce rigorous spatial evaluation metrics (mAP@[.5:.95] with rotated IoU, F1-score across vessel size tiers, false-alarm rates per square kilometer of ocean).4. Optimization for Production & Edge Deployment
Quantize and optimize models (INT8/FP8 using TensorRT, ONNX Runtime, OpenVINO) for high-throughput cloud inference and potential execution aboard SkyServe’s STORM/SURGE onboard edge compute satellite payloads.Package vision inference microservices into lightweight, GPU-accelerated Docker containers integrated with Kafka streaming pipelines.What We Are Looking For
Must-Have Qualifications:
Experience: 3+ years of hands-on experience developing and deploying computer vision / deep learning models in production, with demonstrable experience working on satellite, aerial, or remote sensing imagery.Frameworks & Programming: High proficiency in Python and deep learning frameworks (PyTorch, TorchVision, Hugging Face), accompanied by numerical libraries (NumPy, OpenCV, SciPy).Geospatial Raster Ecosystem: Solid experience handling geospatial raster formats (GeoTIFF, NetCDF, HDF5, SAFE) using libraries such as Rasterio, GDAL, Fiona, Shapely, or Xarray.Object Detection / Segmentation: Deep understanding of modern vision architectures:Rotated/Oriented object detection (e.g., RoI Transformer, oriented R-CNN, YOLO-OBB).Semantic/instance segmentation (Mask R-CNN, UNet, Segment Anything / SAM adaptations for EO).Sensor Modality Knowledge: Firm theoretical and practical grasp of at least one core modality (SAR imaging physics, polarimetry, or optical multi-spectral bands).Nice-to-Have / Strong Advantages:
Direct prior experience with satellite SAR missions (Sentinel-1, ICEYE, Capella, TerraSAR-X, RCM) or EO constellations (Sentinel-2, Planet, WorldView).Familiarity with maritime open datasets: xView3-SAR, OpenSARShip, SSDD, Airbus Ship Detection, or SeaSat.Experience deploying deep learning models on GPU-accelerated edge systems (NVIDIA Jetson) using TensorRT or Triton Inference Server.Publications or open-source contributions in geospatial computer vision venues (CVPR EarthVision, IGARSS, ISPRS).Why Join SkyServe?
Real Strategic Impact: Your models are directly deployed to safeguard marine protected areas, combat forced labor/IUU fishing, and provide actionable intelligence to coast guards and naval defense commands.True Multi-Sensor Problem Space: You will tackle real physical challenges at the boundary of satellite imaging geometry, radar cross-sections, and ocean dynamics.Rapid Deployment Cycle: See your models go from Jupyter notebook prototypes to production cloud platforms and active on-orbit edge compute flights.
