Environmental Sustainability in Practice

Image Classification

The main objective of image classification is to organize each image pixel into a particular class or group based on spectrally similar characteristics. There are several reasons why we might want to classify an image; for example, to make sense of the landscape, and to place a landscape into categories (e.g., forest, urban, rural, wetland, beach, water, etc.).

Classification of land-use and land-cover type is a common data analysis technique in environmental sustainability applications. Often in studies of environmental sustainability issues, we are interested in the natural and human features present on the landscape, and how the landscape has (or may have) changed and evolved over a period of time. In these applications, particular land-use or land-cover classes are identified based on the spectral characteristics of each respective pixel in an image.

There are two main types of image classification: supervised and unsupervised. The decision to use a supervised or unsupervised classification approach may be influenced by several factors, including the spatial and spectral resolution of the image and the complexity of the area under investigation.

Supervised image classification

This classification consists of three stages: training, classification, and output.

1. Training

In the training stage, the image analyst is “training” the computer algorithm by selecting sites in the image that are homogeneous examples of known land-use/land-cover types. These sites are known as training sites or calibration sites because the spectral characteristics of these areas will be used by the computer algorithm to classify the remaining pixels in the image. The selection of training sites is usually informed by ancillary data such as aerial photographs and familiarity with the study site.

2. Classification

In the classification stage, the computer algorithm uses the spectral characteristics of the training sites to classify the pixels in the image.

3. Output

This stage involves the presentation of the results, such as the creation of thematic maps; the results of a classification may also be inputted into a GIS for further investigation and analysis.

Unsupervised Classification

In unsupervised classification, the classes are grouped first by the computer algorithm, based on their spectral characteristics and then the image analyst assigns the classes into thematic classes of interest (e.g., urban, forest, wetland, etc.). A phrase that may be useful to remember the main steps in unsupervised classification is “classify then identify”.  

This page has paths:

This page references: