Image Classification and Accuracy Assessment

This section provides information related to image classification and accuracy assessment. Currently, this section includes information and videos on traditional image classification processes (e.g. supervised classification) and accuracy assessment. As time goes on, I hope to provide information and instructions for Deep Learning.

Image Review and Segmentation

To begin the image classification process, the image (in this case a Landsat 8 image) is reviewed and then the first step, image segmentation, is performed. Review the video below for these steps.

Image Classification Wizard and Selecting Training Samples

The next step (series of steps is to walk through the image classification process using the Image Classification Wizard. After segmenting the image, the next step is to develop a series of training sites (training samples) that contain known land cover classes based on the land cover classification scheme (also part of the wizard).

This video focuses on the training sample manager to choose and name segment samples from the segmentation results.

Perform the Image Classification and Refinement

After the training samples are created and reviewed, the actual image classification process can occur. Upon completing the image classification, a series of “post classification” steps can be used to improve the overall land cover classification before an accuracy assessment is performed. This video completes the image classification wizard process.

Accuracy Assessment

The classification is really not complete without an accuracy assessment. This step performs a quantitative assessment commonly referred to as an Accuracy Assessment, Error Matrix, or Confusion Matrix. These tables are often found in published literature where image classification is performed. This video shows how to complete this step and explains the various components of an accuracy assessment matrix.