Thin Cloud Removal From Remote Sensing Images Using Dual Tree Complex Wavelet Transform And Transfer Support Vector Regression

Remote sensing images which from satellite sensors have the problem of cloud cover problem in different degrees because of climate greatly reduces the utilization rate of remote sensing image, the image post-processing and application. Thin cloud removal approach can effectively improve the utilization and availability of satellite images, and it also can be a very important research in remote sensing images preprocessing.In this paper, thin cloud images from Landsat series and the HJ 1A/B series satellite sensors were selected, using three kinds of effective cloud algorithms, the main research contents include the following aspects:1. Expounded the development of satellite remote sensing technology, Introduced the main parameters of Landsat and the HJ1 A/B series satellite sensors, analyzed the research status of cloud removal algorithm, the dual tree complex wavelet transform and support vector machine theory.2. A thin cloud removal algorithm from remote sensing images based on dual tree complex wavelet transform was proposed. Using multi-player dual tree complex wavelet transform for thin cloud covered remote sensing images, the low frequency coefficient mainly contains the thin cloud information, the high frequency coefficient mainly contains the ground information, effective removal of thin cloud in remote sensing images by high frequency compensation and low frequency suppression and recovery of ground information in cloud coverage areas, Experimental results show that the proposed method can effectively remove the thin cloud and has a fast computing speed.3. A thin cloud removal algorithm for remote sensing image using transfer least squares support vector regression combined with multi-direction dual tree complex wavelet transform was proposed. Based on the dual tree complex wavelet transform, the multiple direction dual tree complex wavelet transform was constructed with the directional filter banks; According to transfer learning theory and least squares support vector regression theory, transfer least squares support vector regression model was proposed, with the help of cloud-free multi-source and multi-phase satellite images of same position and used the transfer least squares support vector regression algorithm to learn the low frequency coefficient of original thin cloud covered image for filling, the original images after multi-direction dual tree complex wavelet decomposition, using adaptive enhancement function enhanced the high frequency coefficients. Clear cloud-free images will be obtained after reconstructed the high frequency coefficients and low frequency coefficients. This algorithm effectively reduced loss of feature information.4. A new algorithm for removing thin cloud from remote sensing images based on multi-direction dual tree complex wavelet transform and transfer twin support vector regression was proposed. According to transfer learning and twin support vector machine theory, transfer twin support vector regression model was proposed. Because the satellite sensor has a certain period of time to get the remote sensing data, the information of the ground objects can be changed inevitably in multi-source multi-temporal remote sensing images that we got. This algorithm used the change detection algorithm based on classification to detect the change of the original images and the multi-source multi-temporal images. Prediction the low frequency coefficients in the region that unchanged by using transfer twin support vector regression model. Then, using dual tree complex wavelet decomposition to the original image that the region had changed. Adaptive enhancement function is adopted to strengthen the high frequency coefficients and suppress low frequency coefficients, the clear cloud-free image can be acquired after reconstruction finally.