Change Detection Metrics And Mixels (mixed Picture Elements) For Computer Analysis Of Low Resolution Remote Sensing Imagery

The computer analysis of low resolution remote sensing imagery such as provided by the LANDSAT I and LANDSAT II satellites presents several basic problems. 1. Resolution elements, which are approximately 80 meters square seldom contain single, identifiable, signatures relevant to land utilization. 2. Picture elements (PIXELS) of mixtures of land usage (MIXELS) often contribute adversely to the success of clustering algorithms and pattern recognition techniques. 3. Since the exact spectral signature (features) of a MIXEL are extremely sensitive to slight displacements of the data, it is difficult to accurately assess changes in basic land usage based on repetitive coverage. This paper discusses several techniques which can be used to find MIXELS and use them in a constructive manner in land use classification. It also presents experimental results obtained from clustering algor ithms for conditions in which MIXELS are both included and excluded from the analysis. In conclusion, the paper suggests several change detection metrics which can be used for gross, low cost screening of LANDSAT data for isolation of regions with changes in land utilization.