Image Classification Lecture Notes

Flashcards covering key terms and definitions related to image classification and remote sensing.

Image classification

The process of identifying the land cover of each pixel in the scene, based on the pixel’s reflectance value in multiple spectral bands.

Land cover

The type of material present on the landscape, such as water, sand, crops, forest, and human-made materials.

Land use

What people do on the land surface, like agriculture, commerce, and settlement.

Information classes

Categories of interest (e.g., geological classes, types of forests) that are not recorded directly on imagery.

Spectral classes

Groups of pixels that are similar regarding their brightness in several bands.

Image space

A 2D-array of pixels containing individual DNs corresponding to reflected or emitted energy within a region of the spectrum.

Feature space

A graph of the N-dimensional vectors formed by the digital numbers (DNs) of a multi-band image.

Clusters

Areas of high density in feature space associated with commonly occurring land-cover types.

Spectral differentiation

The extent to which clusters overlap or are separated due to similar reflectance curves.

Classification scheme

A classification system that includes taxonomically correct definitions of classes organized according to logical criteria.

Parametric classification

Classification methods like maximum likelihood that assume normally distributed remote sensor data.

Non-parametric classification

Methods such as nearest-neighbor classifiers that do not assume normal distribution of remote sensor data.

Hard classification

A classification logic producing discrete categories, such as forest or agriculture.

Fuzzy classification

Classification logic that accounts for the heterogeneous nature of the real world.

Supervised classification

Approach where the identities and locations of some land-cover types are known a priori.

Unsupervised classification

Method where the identities of land-cover types are not known a priori, requiring the computer to group pixels automatically.

k-nearest neighbor (k-NN) classification

A supervised classification method where unclassified pixels are assigned to the most common class among their nearest neighbors.

Fuzzy c-means (FCM) Algorithm

A soft clustering algorithm allowing data points to belong to multiple clusters with varying degrees of membership.

DBSCAN

Density-based spatial clustering method that identifies clusters as high-density regions; does not require a preset number of clusters.

Random Forest (RF)

An ensemble classifier that creates multiple decision trees from random subsets of training data.

Artificial Neural Networks (ANNs)

Classifiers based on brain structure that learn complex patterns in data, but can risk over-fitting.

Object-based image segmentation

Classification approach that segments images into homogenous areas for more meaningful features.

Machine Learning (ML)

A set of algorithms becoming prevalent in image classification, including supervised parametric and non-parametric methods.

Maximum Likelihood Classification (MLC)

A common supervised parametric classifier effective with unimodal data.

Support Vector Machines (SVM)

A machine learning algorithm that identifies optimal hyperplanes to separate data points in multi-dimensional space.