Skip to content

Glossary

This page defines common geospatial and technical terms used throughout the Habitat-Mapper documentation.

Geospatial Terms

Band
A layer of data in an image representing a specific wavelength of light captured by a sensor. For example, separate bands for Red, Green, Blue, and Near-Infrared (NIR) light. Multi-band images allow analysis beyond what the human eye can see.
Coordinate Reference System (CRS)
The geographic framework that defines where on Earth an image is located. Also called a "coordinate system" or "projection". Common examples include WGS84 (used by GPS) and UTM zones.
GeoTIFF
An image file format (.tif) that includes embedded geographic information such as coordinates, projection, and pixel size. This allows GIS software to automatically place the image in the correct location on a map.
NIR (Near-Infrared)
Light with wavelengths just beyond what humans can see (roughly 700-1400 nanometers). Many sensors capture NIR because it's useful for detecting vegetation health and water content. Healthy plants strongly reflect NIR light.
Nodata Value
A special pixel value in a raster that indicates "no information here". For example, black areas outside the survey boundary in a drone orthomosaic are often marked with a nodata value of 0 so software knows to skip them.
Orthomosaic
A geometrically corrected aerial image created by stitching together many overlapping drone or aircraft photos. Distortions from camera angles and terrain are removed, making it suitable for accurate measurements.
Pixel Resolution
The ground distance represented by one pixel in an image, often expressed in centimeters or meters. For example, "5cm resolution" means each pixel covers a 5cm × 5cm area on the ground. Smaller values mean higher detail.
Raster
An image or grid where data is stored as a matrix of pixels (cells). Each pixel has a value representing something like color, elevation, or classification. Satellite images and drone photos are rasters.
Segmentation
The process of classifying each pixel in an image into different categories (e.g., kelp, water, rock). Habitat-Mapper performs segmentation to identify where species are present.
Tiled GeoTIFF
A GeoTIFF organized internally into small square chunks (tiles) rather than as continuous rows. This structure dramatically speeds up reading small portions of large images, which is essential for Habitat-Mapper's sliding window approach.

Technical Terms

Batch Size
The number of image tiles processed simultaneously by the model. Larger batch sizes can improve speed if you have sufficient GPU memory, but may cause crashes if set too high.
CLI (Command Line Interface)
A text-based way to interact with software by typing commands into a terminal, rather than clicking buttons in a graphical interface. The hab command is Habitat-Mapper's CLI.
Crop Size / Tile Size
The dimensions (in pixels) of the square windows Habitat-Mapper uses to process large images. Larger crop sizes reduce stitching artifacts but require more memory. Must be an even number (e.g., 1024, 2048, 3200).
Inference
The process of applying a trained machine learning model to new data to make predictions. When you run hab segment, you're performing inference on your imagery.
Model
A trained deep learning algorithm that has learned to recognize patterns in imagery. Habitat-Mapper includes multiple models for different sensors and species (e.g., kelp-rgb, mussel-gooseneck-rgb).
ONNX
Open Neural Network Exchange, a standard format for saving machine learning models. Habitat-Mapper uses ONNX models so they can run efficiently on different hardware without needing specific deep learning frameworks.
Virtual Environment
An isolated Python installation that keeps Habitat-Mapper's dependencies separate from other software on your computer. This prevents version conflicts and makes installation cleaner.
uint8 / uint16
Data types for storing pixel values as unsigned integers. uint8 uses values 0-255 (8 bits), while uint16 uses 0-65535 (16 bits). Habitat-Mapper expects input images in one of these formats.