Pyropath: An ML Driven Forest Management Path-Planning Solution

Srinidhi Krishnamurthy

Austin, Texas

4 0

Pyropath automates the land management processes used by forestry specialists and first responders in order to prevent catastrophic wildfires and protect wildlife and people - all while saving time, accuracy, cost, and lives. ...learn more

Project status: Published/In Market

Robotics, Artificial Intelligence

Intel Technologies
OpenVINO

Docs/PDFs [1]Code Samples [1]

Overview / Usage

We’ve all heard about the devastating fires that ravaged Australia’s landscapes and made global headlines just a few months back. 46 million acres burned, 3 billion animals dead or displaced, and thousands of people made homeless with 34 fatalities. In the U.S., the Camp Fire in the City of Paradise, CA was the single most destructive wildfire in CA’s history, the worst in America in a century, and the most expensive natural disaster in the world in 2018. 50,000 civilians displaced, 85 fatalities, over $2 billion in damages, 19,000 structures burned, the list goes on.

From wildlife to human lives, government agencies to insurance companies, everyone is affected, and these disastrous, uncontrolled wildfires happen way too often.

We spoke with forestry professionals Len Nielson (Staff Chief of CAL FIRE) and Dr. Richard Harris (40+ years of forestry experience) on some of the pain points they face when maintaining the forests and reducing risk of fire. Current land management practices are:

Time-intensive – Assessing where to put fire treatments can take 6 or more months since most of the work is conducted by foot across acres of forest.
Prone to human error - The processes currently rely on human estimation and judgment, which may lead to inaccuracies/overlooked areas as well as a lot of uncertainty on where to place fire treatments. Furthermore, accuracy may be compromised as human fatigue increases.
Costly – These practices are very expensive since they require so much manual labor, and they are just barely funded by donations from NGOs and government funds.
Infrequent - Because it’s both a time-intensive and costly process, these prevention measures only occur every 5-10 years, allowing for vegetation and forest fuels to grow back and increasing the likelihood of a catastrophic fire. This makes forest escape routes unreliable, as first responders are unable to reach people quickly if the paths are overgrown.

Pyropath eases the pain points of fire professionals and first responders by automating the tree detection and optimal path identification processes used across a wide range of land management methods through aerial view drone or satellite shots. Pyropath is a flexible solution that can be applied to various situations, such as:

Mitigating Dangerous Wildfires - Identifies where to best place fire treatments to mitigate the spread of fires and prevent catastrophic damage to property and wildlife
Preserving Forest Health - Identifies patches of high-density trees for forest thinning applications which break down forest fuels, make forests safer, and allow wildlife to thrive
Keeping Nature-Lovers Safe - Identifies optimal escape routes for civilians, fire specialists, and first responders in wildfires or other time-sensitive situations
Saving Lives Quicker - Identifies the quickest path to an injured, lost, or endangered hiker where every second counts, all while minimizing impact on the environment

Pyropath has the support of fire professionals Len Nielson and Dr. Richard Harris. But beyond the fire specialists, forest rangers, and first responders who will directly be using this solution, Pyropath also benefits hikers, nature-lovers, civilians, wildlife, even insurance agencies. In short, Pyropath benefits everyone.

Methodology / Approach

We are using a custom model architecture called DeepForest in order to identify tree crowns given an input image in a .jpg, .png, or .tif format. This architecture was trained on Ariel California forest images. This identification process yields a bounding box output, which we then convert to a coordinate text file containing numerical representations of where the tree's are. Using this text file, we are able to generate a 2D node map of the trees and populate all empty areas of the map with nodes. Polygons are then used to indicate start and end points on the node map. We then use an A* shortest path algorithm to calculate the shortest path from the starting point to the end point.

OpenVINO was used for deployment of this model though the model optimized and the inference engine. With the power of OpenVINO, PyroPath can be run on any laptop!

Created by Team Pyrogenesis for OpenVINO Challenge 2020.