Data mining in remotely sensed images: a general model and an application

Describes a general data mining model for investigating remotely sensed imagery data. The authors use data mining methodologies to automatically identify patterns (or segmentation classes) and their respective attributes within the image, thus enabling a complete and meaningful description of all pixels. The Data Investigation Model for Unsupervised Segmentation (DIMUS) consists of five modules: inspector, clues generator, classifier, justifier, and mapper. Within the framework of this model, various data mining issues, such as data transformation and selection, information description, determination of segmentation classes, knowledge verification and presentation, are addressed. The authors have applied the model and implemented a fully automated technique that mines remotely sensed images to learn significant classes or patterns through unsupervised clustering. The authors have tested successfully the technique on a variety of imagery domains