Clustering algorithms for aerial photographs and high resolution satellite images

This work, and specially, the use of clustering algorithms was motivated by the need to perform a field-study with erosion data from arid areas. Using data obtained from analyzing erosion, land degradation and desertification phenomena will show some limitations. If only terrestrial observations are considered. Specially, if we are interested in, for instance, forecasting problems of erosion spread. An improvement of the data is possible, if aerial photographs and recent high resolution satellite images are additionally taken into account. The uprising problem with such images is that they contain a huge amount of information, and standard processing algorithms are, in most cases, unable to answer the analyst needs. In order to solve these problems, a compression and suitable selection of the underlying information is needed. Although the development of computer has reached a stage that enables the handling with huge data-sets, considerations concering time complexity are still relevant. In this paper, we present the developed algorithms and discuss possible improvements to attein our aim in performing a classification within a suitable computational time. In section 2, we describe algorithms, such as ISODATA and PHASE, that are based on the classical k-means algorithm. Section 3 describes two ways of finding a set of good starting seeds (centroids) for classification with an adapted method from the known single linkage and the Kohonen networks, as well. Section 4 presents the application of the methods from section 3 to aerial photographs and high resolution satellite images.