Research On Removing Thin Cloud In Single High-resolution Remote Sensing Image

With the development of remote sensing technology, high-resolution remote sensing has been widely used, but vulnerable to the influence of atmospheric clouds. In practical applications, the cloud cover problems also appear in the high-resolution remote sensing images, such as SPOT, IKONOS, QuickBird, etc. Cloud noise not only affects interpretation accuracy of the remote sensing image and reading accuracy of the target features, but also increases the difficulty of obtaining effective remote sensing data and reduces the timeliness of remote sensing when the data comes from temporal resolution lower remote sensing platform. Therefore, it is most significant to effectively reduce or eliminate the influence of cloud on remote sensing image for improving the utilization of remote sensing image.This thesis presents a thorough study of the commonly used methods for removing thin cloud cover from remote sensing based on the study of cloud formation model and related theories, and proposes a new method for removing thin cloud cover from high-resolution remote sensing image based on static wavelet transform. In the end, it takes an experiment using QuickBird images and gets good results. The work of this thesis mainly embodies aspects as follows:1. Studies the commonly used methods existing for removing thin cloud cover from remote sensing images, including cloud homomorphic filter method, Laplace airspace enhancement method, and wavelet transform coefficient weighting method, and points out the advantages and disadvantages of these methods.2. Discusses how to choose the wavelet algorithm, wavelet function, wavelet decomposition level in the wavelet decomposition of image, and analyzes the characteristics of wavelet coefficients, in order to provide theoretical basis for proposing the new method.3. Proposes a new method for removing thin cloud cover from high-resolution remote sensing image based on static wavelet transform. Due to the time-frequency local characteristic and the constant motion characteristic of static wavelet transform, the method firstly establish cloud area thickness mask using low-frequency coefficients, then decreases low-frequency coefficients and increase high-frequency coefficient according to the thickness of mask, finally reconstructs the low-frequency coefficients and high-frequency coefficients processed to get a cloud removal image.4. Introduces some statistical indexes, such as gray mean, standard deviation, entropy, average gradient, correlation coefficient, spectral distortion degree, to evaluate the results of experiments from statistical and visual two aspects. The results show that the method proposed gets better effect than other methods.