Skip to content

Releases

Megalith LiDAR v3.2.0

Choose a tag to compare

@github-actions github-actions released this 06 Nov 02:26

Initial F-Droid release (v3.2.0):

  • Native Android app wrapping the UK 1 m LiDAR & Aerial Cropmark Monument Survey atlas with bundled Leaflet 1.9.4 and offline England candidate dataset.
  • Interactive 7-panel 256 m site inspector (bare-earth slope, 16 m local relief, surface canopy, model probability, 16 m aerial cropmarks, and 1888 OS Six-Inch map cross-checks).
  • Native Android FileProvider sharing for OsmAnd plugin (.osf), filtered GPX waypoints (.gpx), and field survey JSON notes.

Full Changelog: https://github.com/GlassOnTin/megalith-lidar/commits/v3.2.0

v3.1.0 — 9,900 km² Out-of-Sample National Survey & PondConfounderUNet

Choose a tag to compare

@GlassOnTin GlassOnTin released this 24 Oct 08:40

Overview

v3.1.0 expands the out-of-sample multi-domain 1 m LiDAR and 16 m scale-space aerial cropmark/soilmark survey from 1,306 to 2,475 held-out 2 km × 2 km test squares (9,900.0 km², 7.60% of England) and introduces PondConfounderUNet (pond_unet_best.pt) alongside fluvial meander scroll-bar and sunken-basin confounder filters.

Live interactive map and 6-panel LiDAR + aerial cropmark inspector:
https://glassontin.github.io/megalith-lidar/

What's New in v3.1.0

  1. PondConfounderUNet (scripts/train_pond.py, megalith_lidar/scanner.py):
    • Trained an 8-channel (6-channel 1 m LiDAR + 2-channel 16 m scale-space aerial cropmark/soilmark) MegalithUNet on 3,102 256 m × 256 m tiles (1,379 closed OpenStreetMap confounder polygons across ponds/lagoons/water basins, quarries/pits/bunds, and levelled courts/pitches/reservoirs vs 1,723 hard NHLE Scheduled Monument and OSM-verified confounder-free terrain negatives).
    • Evaluated on an untouched 10 km × 10 km BNG spatial block test split (n = 447 tiles, 8-way D4 TTA): ROC-AUC = 0.7971 (95% CI: 0.7535 .. 0.8358), PR-AUC = 0.7262, positive segmentation Dice = 0.2732, and LR+ = 13.12 at the 1% FPR threshold (0.8060).
    • Integrated directly into megalith_lidar/scanner.py, megalith_lidar/osm_context.py, and megalith_lidar/arch_context.py so every candidate is scored for pond_unet_tile_prob, pond_unet_local_peak, and pond_unet_predicted_class.
  2. Sunken-Basin Pond & Fluvial Meander Scroll-Bar Confounder Filters (megalith_lidar/osm_context.py, megalith_lidar/scanner.py):
    • Flags sunken-interior ponds, marl pits, slurry lagoons, and recessed U-bunds (lidar/slurry_lagoon_or_bund) where the interior basin sits at or below the surrounding terrain (centroid_local_relief_m <= +0.05 m, patch_min_relief_m <= -0.28 m) inside a raised perimeter bank.
    • Measures connected ridge continuation span (ridge_continuation_span_m) and adjacent palaeochannel depression span (adjacent_channel_span_m) to flag floodplain meander scroll-bars (lidar/meander_scroll_bar).
  3. Completed 9,900.0 km² Out-of-Sample National Survey (2,475 2 km × 2 km test squares):
    • Prehistoric (candidates.geojson): 589 rediscovered NHLE Scheduled Monuments, 40 rediscovered unscheduled OSM archaeological sites, and 413 unidentified candidates at the strict 1% FPR operating point (1,772 at 5% FPR).
    • Iron Age & Prehistoric Enclosures (enclosure_candidates.geojson): 1,107 rediscovered NHLE Scheduled Monuments, 54 rediscovered OSM archaeological sites, and 53 unidentified enclosure/settlement candidates (20 at the strict 1% FPR operating point).
    • Roman & Romano-British (roman_candidates.geojson): 2,237 rediscovered NHLE Scheduled Monuments, 219 rediscovered OSM archaeological sites, and 1,233 unidentified candidates at the 1% FPR operating point (5,336 at 5% FPR).

Attached Release Assets

  • candidates.geojson — Prehistoric detections (EPSG:4326 geometry with EPSG:27700 BNG coordinates, LiDAR relief, and 16 m cropmark/soilmark metrics)
  • enclosure_candidates.geojson — Iron Age & Prehistoric Enclosure / Hut Circle detections
  • roman_candidates.geojson — Roman & Romano-British detections (with PondConfounderUNet annotations)
  • survey_summary.json, enclosure_summary.json, roman_summary.json — Full 9,900.0 km² survey stream and regional density summaries
  • evaluation_metrics_v2_unbiased.json, enclosure_evaluation_metrics.json, pond_evaluation_report.json — Held-out spatial block test evaluations and 95% bootstrap CIs

v3.0.0 — Multi-Domain 1 m LiDAR & 16 m Aerial Cropmark National Survey (5,224 km²)

Choose a tag to compare

@GlassOnTin GlassOnTin released this 23 Oct 15:53

Overview

v3.0.0 completes the out-of-sample multi-domain 1 m LiDAR and 16 m scale-space aerial cropmark/soilmark survey across 1,306 held-out 2 km × 2 km test squares (5,224.0 km², 4.01% of England).

Live interactive map and 6-panel LiDAR + aerial cropmark inspector:
https://glassontin.github.io/megalith-lidar/

Multi-Domain Detectors & Held-Out Test Benchmarks

All models are evaluated on non-overlapping 20 km × 20 km British National Grid (EPSG:27700) blocks using 8-way $D_4$ dihedral test-time augmentation and validation-locked Platt calibration across five negative null distributions (random_national, matched_environment, anthropogenic_non_target, woodland_canopy_and_pit, trackside_and_field_boundary):

Detector Target Classes Dataset (train / val / test) Held-Out Test ROC-AUC (95% CI) random_national Null ROC-AUC trackside_field Null ROC-AUC Strict 1% FPR Test LR+
Prehistoric (MegalithUNet) Round barrows/cairns, long barrows/chambered tombs, henges/stone circles 4,654 (3,965 / 386 / 303) 0.8675 ([0.8248, 0.9070]) 0.8976 0.8899 61.05× (FPR = 0.63%)
Iron Age & Enclosures (EnclosureUNet v3, 8-Ch) Hut circle & pound settlements, hillforts & banjo enclosures, cross-ridge dykes & coaxial fields 4,323 (3,743 / 315 / 265) 0.8654 ([0.8128, 0.9119]) 0.8941 0.9614 54.76× (FPR = 0.55%)
Roman & Roads (RomanUNet) Roman villas/buildings, Roman road aggers/trackways, forts/camps/settlements 3,928 (3,358 / 282 / 288) 0.7967 ([0.7406, 0.8495]) 0.7741 0.8647 7.00× (FPR = 2.23%)

What's New in v3.0.0

  1. 8-Channel LiDAR + 16 m Scale-Space Aerial Cropmark/Soilmark Pipeline (megalith_lidar/aerial_cropmarks.py, megalith_lidar/model.py):
    • Adds EPSG:27700-aligned 1 m/px vertical aerial orthophoto fetching and zero-mean 16 m scale-space Excess-Green (ΔExG_16m) and luminance (ΔL_16m) cropmark/soilmark channels gated to bare/crop ground (nDSM < 2.0 m).
    • Uses a 7 × 7 LiDAR candidate halo gate and 50% modality dropout so cropmarks boost low-relief ploughed earthworks without triggering false positives on flat tractor tramlines.
    • Enriches all 8,285 published candidate detections across all three domains with quantitative concentric ring-ditch ΔExG and core ΔL contrast metrics (has_distinct_cropmark).
  2. 6-Panel Interactive Diagnostic Inspector (index.html):
    • Adds live browser <canvas> rendering of Panel 5: Aerial Orthophoto (RGB, 256 m × 256 m BNG) and Panel 6: 16 m Scale-Space Cropmark & Soilmark (ΔExG / ΔL) alongside the 4-panel 1 m LiDAR breakdown and an optional Aerial Orthophoto map basemap toggle.
  3. Completed 5,224.0 km² Out-of-Sample National Survey (1,306 2 km × 2 km test squares):
    • Prehistoric (candidates.geojson): 524 rediscovered NHLE Scheduled Monuments, 20 rediscovered unscheduled OSM archaeological sites, and 265 unidentified candidates at the strict 1% FPR operating point (1,135 at 5% FPR; 1,658 / 2,099 with distinct aerial cropmark/soilmark contrast).
    • Iron Age & Prehistoric Enclosures (enclosure_candidates.geojson): 844 rediscovered NHLE Scheduled Monuments, 49 rediscovered OSM archaeological sites, and 35 unidentified enclosure/settlement candidates (17 at the strict 1% FPR operating point; 934 / 1,328 with distinct aerial cropmark/soilmark contrast).
    • Roman & Romano-British (roman_candidates.geojson): 1,109 rediscovered NHLE Scheduled Monuments, 138 rediscovered OSM archaeological sites, and 777 unidentified candidates at the 1% FPR operating point (3,211 at 5% FPR; 3,961 / 4,858 with distinct aerial cropmark/soilmark contrast).

Attached Release Assets

  • candidates.geojson — Prehistoric detections (EPSG:4326 geometry with EPSG:27700 BNG coordinates, LiDAR relief, and 16 m cropmark/soilmark metrics)
  • enclosure_candidates.geojson — Iron Age & Prehistoric Enclosure / Hut Circle detections
  • roman_candidates.geojson — Roman & Romano-British detections
  • survey_summary.json, enclosure_summary.json, roman_summary.json — Full survey stream and regional density summaries
  • evaluation_metrics_v2_unbiased.json, enclosure_evaluation_metrics.json — Stratified bootstrap 95% CIs and subgroup evaluation tables

v1.0.0: OsmAnd Custom Plugin (.osf) & National LiDAR Waypoints (.gpx)

Choose a tag to compare

@GlassOnTin GlassOnTin released this 26 Oct 17:03

OsmAnd Custom Plugin package (megalith-lidar.osf) and standalone GPX waypoints (megalith_lidar_osmand.gpx) for the UK 1 m LiDAR and 16 m aerial cropmark monument survey.

Included in megalith-lidar.osf:

  • 4 OsmAnd Favourite groups (1888 OS Confirmed Antiquities, Prehistoric Candidates, Iron Age & Settlement, Roman Candidates) with relief, walkability notes, and links to the 7-panel web inspector.
  • GPX track for 360 British Roman roads and prehistoric ridgeways (UK_Roman_and_Ancient_Roads.gpx).
  • Online map tile sources for the UK 1888-1913 OS 6-Inch map and the Digital Atlas of the Roman Empire (DARE).

Open megalith-lidar.osf in OsmAnd on Android or iOS to install the plugin, map sources, and monument Favourites.