AstroLens

AI-Powered Astronomical Image Anomaly Detection

v1.1.0 — Released
View on GitHub See the Pipeline

Architecture

End-to-End Intelligent Pipeline

From raw sky survey data to actionable discovery reports — fully autonomous, self-correcting, and production-ready.

AstroLens end-to-end ML pipeline

Results

What AstroLens Found

Autonomous analysis across multiple sky surveys with zero human intervention.

22,195
Images Analyzed
3,541
Anomaly Candidates
99.5%
YOLO Detection Accuracy
5,471
Unique Sky Regions
0
Human Intervention Required
269
Known Objects Recovered
Including SN 2014J, NGC 3690, SDSS J0252+0039

Demo

See It in Action

AstroLens provides an intuitive web interface for detection, gallery viewing, and result visualization.

AstroLens anomaly detection in action

Anomaly Detection

AstroLens gallery and viewer

Gallery & Viewer

AstroLens visualization and analysis

Visualization & Analysis

Technology

How It Works

An end-to-end intelligent pipeline from raw sky survey data to actionable discovery reports.

Download from Sky Surveys
ViT Classification
OOD Anomaly Detection
YOLO Transient Detection
Catalog Cross-Reference
Discovery Report
Python PyTorch FastAPI YOLOv8 Docker

Runs on laptop: CPU, Apple MPS, and NVIDIA CUDA supported

Infrastructure

Cloud Architecture on AWS

Production deployment patterns that take AstroLens from local research into cloud-native, scalable infrastructure.

ML Inference Pipeline

Managed services orchestrate the full ML pipeline at scale on AWS.

ML Inference Pipeline - AWS Production Architecture

Infrastructure Architecture

Full AWS platform view — networking, compute, storage, and security layers.

Infrastructure Architecture - AWS Platform View

DevOps Pipeline

End-to-end CI/CD and deployment automation from commit to production.

CI/CD and Deployment Pipeline

AWS Services

Managed services powering the production ML pipeline.

Amazon SageMaker
Model training, fine-tuning, and managed inference endpoints
AWS Step Functions
Orchestration of multi-stage ML pipelines (download, classify, detect, cross-reference, report)
Amazon Bedrock
Foundation model integration for enhanced classification and natural language analysis
Amazon S3
Data lake for survey images, model artifacts, and results
Amazon SQS / SNS
Event-driven processing and alert notifications
AWS Lambda
Serverless preprocessing, image validation, and lightweight inference
Amazon DynamoDB
Low-latency metadata store for candidates and cross-reference results
Amazon CloudWatch
Monitoring, alerting, and pipeline health dashboards
Amazon ECR + ECS/EKS
Container orchestration for scalable inference workers

Data Flow: Inputs, Processing, Outputs

Inputs
  • Sky survey image archives (SDSS, ZTF, DECaLS)
  • Transient alert streams (TNS, ZTF alerts)
  • Astronomical catalog data (SIMBAD, NED)
  • Pre-trained model artifacts (ViT, YOLO)
Processing
  • Step Functions orchestrates multi-stage pipeline
  • Lambda validates and preprocesses images
  • SageMaker runs ViT + OOD + YOLO inference
  • Bedrock provides enhanced analysis context
  • Cross-reference against catalog databases
Outputs
  • Ranked anomaly candidates with confidence
  • Cross-reference verification status
  • Morphology and classification reports
  • Real-time SNS alerts for high-confidence detections
  • CloudWatch dashboards and operational metrics

DevOps & Platform Engineering

End-to-end delivery covering design, implementation, automation, and operations.

Infrastructure as Code

Terraform, CloudFormation, CDK — repeatable, version-controlled infrastructure

CI/CD Pipelines

GitHub Actions, CodePipeline — automated build, test, and blue-green deployment

GitOps & ArgoCD

Declarative deployments, drift detection, and rollback from a single source of truth

Container Orchestration

ECS and EKS workload management, auto-scaling, service mesh, multi-AZ resilience

Security & Compliance

IAM policies, GuardDuty, Security Hub, automated vulnerability scanning in pipeline

Observability

CloudWatch, X-Ray, Prometheus, Grafana — full-stack monitoring and distributed tracing

Cost Optimisation

Right-sizing, Savings Plans, Spot instances, Cost Explorer dashboards

Now Live

MitraSETI Cloud — Start Free

AstroLens ML models power the classification engine in MitraSETI Cloud. Upload observation data, run automated signal detection and ML classification pipelines, and visualise results — directly in your browser.

☁

Upload & Process

Upload .h5, .fil, .fits, and .hdf5 files up to 1 GB. Automated pipelines handle the rest.

📊

ML Classification

CNN + Transformer models classify signals and flag anomalies automatically.

🔗

Interactive Visualisation

Waterfall plots, signal galleries, and pipeline monitoring — real-time in your browser.

Try MitraSETI Cloud — Free Explorer Tier

Free tier available • No credit card required • Researcher and Institution tiers for larger workloads

Open Source

Explore the Code

AstroLens is fully open source. Clone, run, and contribute — science advances fastest when tools are transparent and accessible.

View AstroLens on GitHub