Joint rectification of image series classification results based on trajectory analysis

Abstract

There has been a common method to study land use changes based on series of remote sensing data. However, data on time nodes in the time series are almost the historical data, which is difficult in acquiring sample mark data for classification and result verification. Thus, the classification accuracy is severely limited. Taking a series of 4 remote sensing imageries of Xihe watershed, Gansu Province, Northwestern China as an example, the authors proposed a new method to jointly rectify image series classification results based on trajectory analysis. The object-oriented classification method was used to classify the remote sensing images, and results were output as vector data, which were then utilized to take trajectory analysis on the change process of every specific point in the study area. The olassification results were rectified by further investigation or expert querying on those patches with unreasonable trajectories. After joint rectification, compared with those with no joint rectification, the classification accuracy for the former three time node improved about 3%-8%, especially for the middle two periods of historical data, the classification accuracy improved up to 7% -8%.