AI-Automated Feature Extraction from Aerial Imagery

Artificial Intelligence is being used to effectively to map vegetation, buildings, utilities, and water bodies.

A New Perspective on New Zealand’s Landscapes

When we monitor urban expansion to protecting our unique ecosystems, we require accurate, large-scale, and up-to-date spatial information. Traditional ground surveys provide essential accuracy, and they are often limited in scope and frequency.

Wellington city captured with a digital twin overlay, identifying various infrastructure, buildings, and road networks which have been identified by AI.

When you train AI models to recognise specific objects like buildings, vegetation, and waterways, we can move from slow, manual analysis to rapid, automated insight. It allows us a more automated and proactive approach to urban planning, environmental stewardship, and infrastructure management across Aotearoa.

Teaching AI Machines to Interpret Images

The technology we are talking about uses a form of AI called deep learning. Specifically, models known as Convolutional Neural Networks (CNNs) are trained to analyse images pixel by pixel. Through a process of being shown thousands of labelled examples, these models learn to identify the unique patterns, textures, and spectral signatures associated with different features on the ground.

This process is often referred to as semantic segmentation, where every pixel in an image is assigned a category, such as ‘building’, ‘road’, ‘water’, or ‘forest’. The result is a highly detailed thematic map, extracted automatically and at a scale previously unimaginable.

Hyperspectral + InSar monitoring of Fiordes in the South Island of New Zealand

Applications of Automated Feature Extraction with AI

Mapping Buildings and Urban Infrastructure

For our growing urban centres, AI models can rapidly process high-resolution aerial imagery to produce precise outlines of every building footprint. Beyond planning, this data supports infrastructure management, solar panel potential assessment, and emergency response planning by providing a clear, current inventory of the built environment. In New Zealand, AI is already being applied to create national datasets of buildings and other features, forming a foundational layer for countless GIS applications. Check out our LandSure InSAR Land Stability Reports Here

AI Analysis of Vegetation and Ecosystem Health

Understanding our vegetation is super important. Things like assessing pasture productivity in our agricultural sector, or monitoring the health of native forests. AI, when combined with multispectral and hyperspectral satellite imagery, provides profound insights.

These advanced sensors capture light in narrow bands beyond the visible spectrum. This reveals detailed information about plant health, moisture content, and species type. AI algorithms are uniquely capable of decoding this complex data to classify different types of vegetation, detect plant stress or disease, and monitor for the spread of invasive species like wilding pines.

AI-generated topographic analysis of the Coromandel Peninsula showing vegetation mapping and land boundaries across native coastal forest.

Interactive Digital Twin Environmental Models

The integration of technologies brings together the concept of a "digital twin"... A dynamic, virtual replica of a physical area that is continuously updated with real-time data. This is more than a static 3D model, its more like a living, evolving system that mirrors the real world that it monitors.

A digital twin for an area like Wellington, for example, can integrate ongoing satellite imagery, live sensor feeds (e.g., from water quality or traffic monitors), and updated infrastructure data. Within this virtual environment, an analytical engine powered by AI can:

• Automatically detect changes, such as new construction or vegetation loss.

• Run simulations to forecast the impact of future events, like stormwater runoff from a new subdivision.

• Flag anomalies that require attention, such as a decline in forest health or unusual ground movement.

This creates a powerful platform for any analyst, to support strong evidence-based decision-making, and allowing planners and managers to test scenarios and understand complex environmental interactions before committing to action on the ground.

Flood Mapping with various data sets can train AI models to forecast for farmers more accurately.

Practical Considerations and The Path Forward

Implementing this technology effectively and accurately, requires a coordinated effort and specific capabilities:

• Robust Data Infrastructure: The large files associated with LiDAR and high-resolution imagery demand significant cloud computing power and sophisticated data management workflows.

  
  

• Interdisciplinary Collaboration: Effective outcomes depend on close collaboration between GIS specialists, data scientists, engineers, and subject-matter experts in fields like urban planning and ecology.

  
  

• Strategic Investment: While access to some satellite data is free, investment is required for high-resolution commercial imagery, specialised data acquisition (like LiDAR), and the development of tailored AI models.

New Zealand CAN harness AI-driven feature extraction to build a comprehensive, near-real-time understanding of our nation. It empowers all of us to manage our growth sustainably, protect our invaluable natural heritage, and build a more resilient future.

If you are interested in learning more, the team at AVANT are some of the world’s leading experts in delivering these technologies locally in New Zealand.