Classification of Landsat Images Based on KLT with Multiple Thresholds

摘要

This paper proposes an automatic multispectral image classification approach based on the Karhunen–Loève transform with multiple thresholds. We used a Landsat ETM 7 image taken around Brasilia city, working with three spectral bands. We distinguished five classes (water, forest, reforestation, bulky field, and field) and plotted a map using all classifications with different thresholds. Our approach performs dimensionality reduction using only eigenvectors with the highest eigenvalues, generating an eigenspace of low dimension. The target recognition is carried out by finding out the shortest Euclidean distance among the primitives of the new images and the primitives of the classes we are working with.