Classification Accuracy Improvement using Wavelet-PCA Image Fusion Technique and Change Detection in Land Cover/Land Use

Abstract

In general, to detect land use land cover (LULC) changes, LULC features are extracted independently using multi-temporal remote sensing data, then analyzed and compared to detect their changes. This study introduces a new classification method that integrates PCA-Wavelet fusion technique and Maximum likelihood classifier (MLC). PCA-wavelet fusion technique was used to extract the quality information about the targets whereas MLC was applied for the purpose of image classification. The quality of fused image was evaluated by the existing fusion methods. MLC classification was applied to extract four classes; settlement land, vegetation, water and soil, for the proposed and existing fusion techniques. Experimental results show that the wavelet-PCA fusion technique improves classification accuracy better than the other existing fusion methods. The overall accuracy and kappa coefficient for the proposed is 0.91 and 0.86. The performance of proposed method was evaluated using Landsat 8 data set. Furthermore, change in each class, in the interval 1978 - 2018, was evaluated with post classification comparison method. Changes were observed in all classes; the results demonstrate that during the forty years period, area under settlements and soil/bare land increased and the area under water bodies and vegetation decreased. In the future, these changes may have a significant influence on the ecosystem.