kbhujbal

kbhujbal / NASA_exoplanet_detection_using_CNN_transfromer

Public

Hybrid CNN-Transformer model for automated exoplanet transit detection on NASA Kepler light curves. Features dual scale CNN branches, Transformer based sequence modelling, Grad CAM + SHAP + attention explainability, and Monte Carlo Dropout uncertainty quantification. 5 fold cross validated with baseline comparisons.

astrophysics convolutional-neural-network deep-learning exoplanet-detection explainable-ai
22
0
100% credibility
Found May 03, 2026 at 22 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

A complete machine learning toolkit for classifying potential exoplanets from NASA's Kepler mission data using a hybrid neural network model with built-in explanations and uncertainty estimates.

How It Works

1
🔭 Discover the Planet Hunter

You find a free tool that uses NASA's telescope data to spot real planets around distant stars by learning patterns in their light.

2
📥 Gather Star Data

Download the public list of potential planet signals from NASA's Kepler mission and place it in your data folder.

3
🛠️ Prepare Your Workspace

Run a quick setup to create organized folders for your data, models, and pictures of results.

4
🚀 Train the Detector

Start the training process where the tool studies thousands of star light curves to learn the difference between true planets and lookalikes.

5
📊 Review Results

Check performance charts, comparisons to simpler methods, and see how accurately it identifies planets.

6
🌟 Test a Specific Star

Pick any star from the list and get an instant prediction with confidence level and visual explanations.

🎉 Confident Discoveries

Celebrate having a reliable planet detector that flags real exoplanets, explains its reasoning, and highlights uncertain cases for double-checking.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 22 to 22 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is NASA_exoplanet_detection_using_CNN_transfromer?

This Python project delivers an automated pipeline for detecting exoplanets in NASA Kepler light curves using a hybrid CNN transformer model. It processes phase-folded data through dual-scale CNN branches and Transformer sequence modeling with attention, outputting planet probabilities plus explainability visuals like attention maps, Grad-CAM heatmaps, and SHAP values. Users get CLI tools for training (train.py), evaluation with baselines (evaluate.py), and single-KOI predictions (predict.py --koi_id K00001.01), all generating plots and JSON results from the Kaggle dataset.

Why is it gaining traction?

It stands out among hybrid CNN transformer architectures by bundling uncertainty via Monte Carlo Dropout, flagging low-confidence cases for review, and providing publication-ready comparisons to Random Forest, LSTM, and CNN-LSTM baselines under 5-fold CV. Devs dig the end-to-end reproducibility—one command preprocesses data, trains, and spits out ROC/PR curves—plus integrated explainability that audits predictions without extra setup. Like other GitHub hybrid CNN transformers for tasks from deraining to tooth detection, it hooks astro ML folks with zero-boilerplate XAI.

Who should use this?

Astrophysicists vetting Kepler candidates tired of manual transit checks. ML engineers prototyping hybrid CNN transformer models on imbalanced time series like light curves. Researchers in automated detection needing uncertainty and attention visuals for papers.

Verdict

Solid prototype for exoplanet workflows with strong docs, a walkthrough notebook, and GPU/CPU support, but 22 stars and 1.0% credibility signal early-stage maturity—run your own benchmarks before production. Worth forking if you're into hybrid CNN transformers.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.