Remote Sensing Image De-thin Cloud Method Based On Daubechies Wavelet Transform

With the rapid development of satellite technology,remote sensing images are widely used in urban planning,dynamic monitoring,weather forecasting,and feature recognition.However,the optical sensor is easily affected by the cloud layer when collecting data,which causes the image object to be occluded,the sharpness is lowered,and the later object recognition is difficult.Therefore,in order to improve the utilization rate of remote sensing data and achieve economical cost savings,it is a popular in remote sensing digital image processing technology to study how to effectively remove image cloud noise.This paper introduces the imaging model of the remote sensing image thin cloud and the basic theory of the cloud removal method.Aiming at the problem that the traditional de-thin cloud method loses a large amount of source information when filtering cloud noise,an improved wavelet transform to thin cloud image method is proposed.The algorithm is obtained by merging the homomorphic filtering method and the high frequency emphasis filter into the traditional wavelet transform.Firstly,by selecting the optimal wavelet base db4 and 3 layer boundary layer;Secondly,the experimental image is decomposed by Mallat algorithm,and the high-level detail coefficient including most cloud noise,the low-level detail coefficient of a large amount of object information,a small amount of cloud noise and the approximation coefficient of background information are obtained.Then,the low-frequency components reconstructed from the high-level detail coefficients are pre-processed by homomorphic filtering and then appropriately increased.The high-frequency components obtained by reconstructing the low-level detail coefficients are increased.High-frequency emphasis filtering processing on the highest layer approximation coefficient.Finally,the wavelet reconstruction results in the de-cloud,which improves the shortcomings of the traditional wavelet transform method in dealing with low-frequency information.In order to verify the effectiveness of the improved algorithm,the single-view Landsat8 image was used as the research object,and the three methods were used to perform de-thin cloud processing on the experimental image.The visual and evaluation indicators and the objective indicators were used to independently and comprehendsively evaluate the de-cloud effect.The experimental results show that:(1)The improved Butterworth high-pass filter can better preserve the details of the original image and avoid the problem of “ringing” and distortion.The homomorphic filtering method can remove the thin cloud with uniform distribution of medium thickness.As the cutoff frequency increases,more cloud noise is removed,but the source information is more lost and the resulting image is more blurred.(2)The traditional wavelet transform method reduces the high-level detail coefficient and the highest layer approximation coefficient,and increases the low-level detail coefficient,which can weaken the cloud noise to a certain extent and enhance the ground object information,but cannot completely filter out the cloud on the low-frequency component.Therefore,the ideal cloud removal effect cannot be achieved.(3)The improved algorithm in this paper can ensure the removal of thin cloud noise more effectively with the least source information loss,and enhance the details and contours of the features,with better contrast and higher definition,and the effect after cloud processing is better.The time domain of the wavelet transform and the frequency domain complementarity of the Fourier transform are realized.Therefore,the improved algorithm in this paper is superior to other methods in de-clouding effect,which verifies the effectiveness of the algorithm.