Workflow Automation of Tree Detection Visualization Comparison(Deep Forest Model)

Introduction

This is an automatic workflow to complete from the first step importing aerial imagery to the last step generating reports for the whole process of comparing tree detection results of DeepForest model (Weinstein et al., 2020). To run the workflow please follow 8 steps in ObjectDetectVisualComp to generate a web comparison report.

Method

• Frontend: Streamlit + html
• Backend: Postgresql (PostGIS)
• Geoprocess: Overlap, Add attribute, Convert to Geojson

Project Folder

• data
  • TreeAOIWGS84.tif
  • Extract_TreeLINZ_03m.tif
• Output
  • run1_predictions.geojson
  • run2_predictions.geojson
  • run1_predictions.csv
  • run2_predictions.cvs
  • settings.csv
  • ComparisonReport.pdf
• run_tile.py
• settingGUI.py
• ObjectDetectVisualComp.ipynb
• ComparisonWebpage.html

Results

1. Model Setting GUI

2. Comparison Report

3. Settings.csv (Relational Table)

	patch_size	patch_overlap	score_threshold	iou_threshold	batch_size	file_name
Run 1	1200	0.25	0.2	0.15	4	run1_predictions
Run 2	800	0.25	0.4	0.15	4	run2_predictions

4. Geojson file attribute Table

xmin	ymin	xmax	ymax	label	score	image_path	geometry
4733	1802	4799	1876	Tree	0.560285925865173	TreeAOIWGS84.tif	POLYGON ((4799 1802, 4799 1876, 4733 1876, 4733 1802, 4799 1802))
4532	2385	4775	2608	Tree	0.53722459077835	TreeAOIWGS84.tif	POLYGON ((4775 2385, 4775 2608, 4532 2608, 4532 2385, 4775 2385))
...

Demo