News Feed

Vegetation health assessment is used in a wide range of environmental assessment programs such as spill clean-up verification and annual vegetation health assessment at risk-managed contaminated sites.

Timely cost-effective assessments using traditional (boots-on-the-ground) field assessment year after year can be difficult (and subjective) due to high variances in vegetation growth rates and species influenced by meteorological conditions, diverse soil types, and topography. This presentation provides a summary of a proof-of-concept demonstration project that uses the power of artificial intelligence to complete vegetation health assessments. The demonstration project exploits the abilities of machine learning computer scripts to independently recognize, identify, sort, and classify complex patterns.

For this initiative, a supervised Machine Learning System (MLS) is trained with a known dataset of vegetation, examples of photos that correspond to vegetation feature classes that require identification and classification. The training dataset is created by subject matter experts who use ground-truth vegetation assessment data to ensure that the training dataset inputted to MLS is accurate. In addition, unsupervised MLS classification analyses are completed and compared against the supervised MLS classification outputs to minimize the introduction of human bias.

MLS scans ultra-high-resolution visible wavelengths of georeferenced air-photos obtained from low altitude Unmanned Aerial Vehicle (UAV) flights to identify and classify the desired vegetation features by matching the vegetation patterns in the UAV photo with the training dataset. MLS is programmed to automatically scan and classify billions of pinpoint locations before it finalizes its prediction. The MLS automation requires little human intervention, enabling it to review air-photo(s) covering areas ranging from thousands of square metres (UAV photo) to millions of square kilometres (satellite photo).

This presentation will also cover how artificial neural network technology can be used to leverage the information from the MLS and site-specific environmental investigation data to solve perplexing real-world problems.