Monitoring Urban Forests from Auto-Generated Segmentation Maps

TL;DR

This paper introduces a weakly-supervised approach to monitor urban forests using remotely sensed data, leveraging LiDAR-generated noisy labels to train segmentation models with minimal human input, demonstrated through Hurricane Sandy's impact assessment in NYC.

Contribution

It proposes a novel weakly-supervised methodology utilizing LiDAR data as noisy labels for urban forest segmentation from orthophotos, reducing the need for manual annotations.

Findings

Effective localization of urban trees using noisy LiDAR labels

Successful assessment of hurricane impact on NYC urban forests

Minimal human interaction required for model training

Abstract

We present and evaluate a weakly-supervised methodology to quantify the spatio-temporal distribution of urban forests based on remotely sensed data with close-to-zero human interaction. Successfully training machine learning models for semantic segmentation typically depends on the availability of high-quality labels. We evaluate the benefit of high-resolution, three-dimensional point cloud data (LiDAR) as source of noisy labels in order to train models for the localization of trees in orthophotos. As proof of concept we sense Hurricane Sandy's impact on urban forests in Coney Island, New York City (NYC) and reference it to less impacted urban space in Brooklyn, NYC.