Remote sensing change detection study based on adaptive threshold in pixel ratio method

This paper mainly proposes a change detection method for different time remote sensing image by combining mean pixel ratio and post-classification comparison. Mean pixel ratio method can get more continuous result comparing with traditional pixel ratio, but the threshold needed is still determined by training sample. The distribution and numbers of sample can have an effect on the value of threshold and further lead to different results of change detection. To solve this problem, we propose an automatic, adaptive threshold determination method that the entire image is evenly sampled, and the appropriate threshold is determined by the histogram method without human intervention. For post-classification comparison, we use supervised classification module in Erdas software to classify two different time images and compare the difference. Our method weights the results of adaptive mean pixel ratio and post-classification comparison. Experiments show that the adaptive threshold determination can ensure the objectivity of threshold and improve the efficiency of change detection and the fusion of the result of two methods can improve the reliability of change detection.