Improved Fast-ICA for Change Detection of Multi temporal Remote Sensing Images

When the selection of the initial points in the Fast-Independent Component Analysis (Fast-ICA) algorithm is far from the minimum point, the problem of convergence may not be convergent, and an improved Fast-ICA algorithm based on the damped Newton method is proposed. The algorithm adds one dimension search along the direction of Newton iteration, thus ensuring the convergence of Fast-ICA algorithm. Secondly, the Contourlet transform is combined with the improved Fast-ICA. When using the combined method for change detection, the initial transformation of remote sensing image is first used by Contourlet transform; secondly, the modified Fast-ICA algorithm is used to detect the change; then the image after the change detection is invert with Contourlet; finally, the image is segmented by threshold value, and the change results are obtained. The experimental results show that, compared with the change detection results based on the traditional Fast-ICA algorithm, the proposed method can effectively separate the information of the change, and the detection precision is higher.