Semi-Supervised Two Stage Classification Technique

摘要

A Semi-Supervised Two Stage Classification Technique has been developed on the IBM PC-AT computer at the Environmental Remote Sensing Center, University of Wisconsin-Madison. This technique is used to classify multispectral digital images. It involves two stages. The first stage is a hybrid clustering technique and the second is a reclassification (post-classification process) of a spectrally classified image with digital ancillary information. In the first stage, the analyst directs the clustering algorithm by delineating a certain number of training areas so that an unsupervised clustering algorithm can identify a user defined number of spectral clusters in each area. In the second stage, ancillary data is employed as a Second Stage of digital information to reclassify certain spectrally classified land cover types to increase the classification accuracy. A SPOT satellite sub-scene over the Greater-Madison area in Wisconsin is segmented utilizing the Semi-Supervised clustering approach. The FINDSET algorithm is an unsupervised clustering algorithm that is presently employed at the Environmental Remote Sensing Center. A comparison between the Semi-Supervised approach and the FINDSET algorithm is assessed.