π Functions & Usage
The Overview tab provides a high-level summary of the geological sample catalog. It displays key statistics (total samples, types, categories, best model accuracy, Cohen's ΞΊ), an interactive global map of Venus showing sample locations color-coded by geological category, and a scientific summary of the dataset and ML pipeline. Use the SAR Background toggle to overlay sample positions on the Magellan radar mosaic. Click legend items to show/hide categories. Zoom and pan the map for detailed exploration.
π For Students & General Audience
This page shows a map of Venus with colored dots β each dot represents a place on Venus where scientists have identified a specific geological feature (like a volcano, crater, or tectonic zone). The numbers at the top tell you how many samples we have and how well our computer model can recognize them. Think of it as a 'bird's-eye view' of our entire dataset.
Scientific Summary
π Functions & Usage
The Data Explorer tab lets you browse, filter, and sort all geological samples in a paginated table. Filter by category, type, or search by name. Click any row to open a detailed side panel showing the SAR patch image, extracted texture features with visual bars, ML prediction with confidence score, SHAP feature contributions, and nearest neighbors in feature space. Export filtered results as CSV for external analysis.
π For Students & General Audience
This is like a searchable spreadsheet of all Venus surface samples. You can filter by type (e.g., only volcanoes) and click on any sample to see its radar image, what the AI thinks it is, and why. The colored bars show different measurable properties of each surface patch β like how rough or smooth it looks in the radar image.
π Functions & Usage
The Analysis tab offers exploratory data analysis tools: category overview cards with sample counts and thumbnails, PCA and UMAP dimensionality reduction scatter plots to visualize how categories cluster in feature space, Fisher discriminant separability analysis showing which feature pairs best separate categories, and a pairwise distance matrix heatmap. These tools help assess whether the texture features provide meaningful discrimination between geological categories.
π For Students & General Audience
This tab helps you see if our measurements can actually tell different geological types apart. The scatter plots show each sample as a dot β if dots of the same color cluster together, it means the computer can distinguish those types. The separability chart shows which measurements are most useful for telling categories apart, like how 'roughness' might separate volcanic areas from smooth plains.
Category Overview
Feature Space Visualization
Separability Analysis
Pairwise Distance Matrix
π Functions & Usage
The Spatial Prediction tab runs sliding-window inference across any Venus region. Select preset regions or define custom coordinates, choose resolution (16/32/64px steps), model type, and single- or multi-scale mode. The output is a geological prediction heatmap with layers for dominant category, sub-type (2-level hierarchical prediction), uncertainty (entropy), category probability, and scale disagreement. Post-processing options include edge-aware smoothing, Gaussian smoothing, and majority vote. High-confidence unlabeled regions are flagged as candidates for expert review. Results can be exported as GeoJSON polygons or prediction grid CSV.
π For Students & General Audience
Imagine sliding a magnifying glass across Venus's surface β at each position, the AI predicts what geological type it sees. The result is a color-coded map showing what the AI thinks the surface is made of. Brighter colors mean the AI is more confident. You can pick different regions of Venus to explore and adjust how detailed the analysis should be. Areas where the AI is very confident but we don't have labels might be interesting places for scientists to study further.
Region Selection
3D Terrain Visualization
π Functions & Usage
The Classification tab presents ML model comparison results: accuracy, F1 scores, and Cohen's ΞΊ for each model (Dummy, Logistic Regression, Random Forest, SVM, XGBoost). It includes confusion matrices (absolute and normalized), per-class precision/recall/F1, SHAP global and per-category feature importance, permutation importance validation, spatial cross-validation for generalization testing, and an optional CNN patch classifier with Grad-CAM visualization. The methodology panel documents the full pipeline.
π For Students & General Audience
Here we compare different AI models to see which one is best at recognizing Venus geological features from radar images. The confusion matrix shows where the AI gets confused (e.g., mixing up volcanic plains with tectonic zones). SHAP charts explain which image properties the AI relies on most β think of it as asking the AI 'why did you make this decision?' The CNN is a deep learning approach that learns directly from images rather than pre-computed measurements.
Model Comparison
Confusion Matrix
Per-Class Metrics
SHAP Feature Importance
Per-Category SHAP Breakdown
SHAP vs Permutation Importance Validation
Spatial Validation
Spatial cross-validation partitions the planet into non-overlapping geographic regions with a 5Β° buffer zone between folds to prevent spatial leakage. Note: on planetary surfaces with strong regional heterogeneity, spatial CV F1 values of 15-30% are typical and expected β the test measures genuine cross-region generalization, not model failure.
CNN Patch Classifier
Train a lightweight convolutional neural network as a morphology-sensitive alternative to hand-crafted texture features. The CNN processes CLAHE-enhanced patches scaled to 64Γ64px (~800K parameters). Estimated training time: ~5-10 minutes on CPU.
▼ Methodology & Limitations
π Functions & Usage
The Venus Atlas tab integrates the IAU Gazetteer of named planetary features. Browse ~200 officially named Venus surface features (craters, coronae, tesserae, etc.) with their coordinates, diameters, and approval dates. Switch between map view and table view, filter by type or diameter, search by name, and click any feature to see its SAR image patch extracted from the Magellan mosaic. Jump directly to Spatial Prediction for any feature's region.
π For Students & General Audience
Venus has hundreds of named features β craters named after famous women, volcanic structures called 'coronae,' and more. This tab lets you explore them on a map or in a list. Click on any named feature to see what it looks like in radar, learn about its size and location, and even run the AI prediction on the surrounding area. It's like having a guided tour of Venus's most interesting landmarks!
Initialize Venus Atlas
Load the database of ~200 named Venus surface features with SAR imagery patches. This requires the base image to be loaded.