Soft classification and land cover mapping from remotely sensed imagery

Although soft classification analyses can reduce problems such as those associated with mixed pixels their accuracy is often low. The key aim of this research is to investigate the ways to increase the accuracy of soft classification, the factors that impact on soft classification and its implications for the real world applications. Four possible methods for combining soft classifications to increase classification accuracy were assessed. All four ensemble approaches were found to increase classification accuracy. Relative to the most accurate individual classification, the increases in overall accuracy derived ranged from 2.20% to 4.45%, increases that were statistically significant at 95% level of confidence. The impact of intra-class spectral variation on the estimation of sub-pixel class composition was investigated. Results showed that the nature of intra-class variation has a negative impact on the accuracy of sub-pixel estimation as it opposed the assumption that a class can be represented by a single spectral end-member. It was suggested that a distribution of possible class compositions could be derived from pixels instead of a single class composition prediction. The impacts of intra-class spectral variation on the use of soft classification for super-resolution mapping were assessed. It was apparent that the accuracy of the soft classification and super-resolution mapping declined as the degree of intra-class variation increased. Finally, a possible method to reduce the impacts of intra-class spectral variation on sub-pixel classification was investigated.