Performance Comparison of Multi-temporal Landsat Remote Sensing Image Relative Radiation Normalization Method

The radiation normalization operation of multi-temporal remote sensing images is an indispensable step before land cover change detection and image stitching. This study was based on the Landsat 8 OLI data covering Nanjing on July 10, 2013 and March 28, 2016, with 2016 images as reference images. The distribution-based histogram matching method and sequential conversion method were used to perform relative radiation normalization operation on images based on pixel-based multivariate change detection method and random forest method. The performance of different relative radiation normalization methods was evaluated by using 5 objective evaluation indexes: information entropy, edge intensity, spatial frequency, peak signal-to-noise ratio and interactive information. The results show that after 4 kinds of normalization methods, it can be seen through visual observation that the image spatial information remains intact and there is no spectral feature of the damaged features. Combined with 5 evaluations and comparisons, it is concluded that the normalization effect of the sequential transformation method is optimal. The research conclusions can provide reference for the collaborative use of multi-temporal remote sensing images.