Research On Preprocessing And Analysis Methods For Remote Sensing Images

With the rapid development of remote sensing technology,it provides reliable and efficient data security for human to realize the landscape and landform of the earth's surface,through the real time images of ground objects by means of satellite and airborne remote sensing technology.Thus research on remote sensing image preprocessing and analysis methods has important theoretical meanings and practical application value.Based on the previous work,the paper studies several key technologies involved in the current field of remote sensing image preprocessing and analysis in-depth,including remote sensing image enhancement,remote sensing image segmentation and edge detection,remote sensing image matching,and remote sensing image fusion.The main work is as follows.Firstly,a remote sensing image enhancement method in non-subsampled shearlet transform(NSST)domain is researched based on guided filtering and artificial bee colony optimization.Firstly,the NSST is performed to decompose the remote sensing image into several high-frequency components and a low-frequency component.Then the high-frequency components are transformed by a nonlinear gain function,through which the edges and details are enhanced and the noise is restrained.The guided filtering is utilized to enhance the low-frequency component avoiding amplifying noise in the process of image enhancement.Finally,the result image is reconstructed from the processed high-frequency components and low-frequency component.Considering that the box filter radius and regularization parameter of guided filtering have significant influences on enhancement effects,the artificial bee colony optimization algorithm is adopted to search their optimal values for best enhancement effects.A large number of experimental results demonstrate that the proposed method has outstanding performance in improving visual effects and quantitative evaluation indicators such as definition and contrast.Compared with four enhancement methods put forward in recent years,the proposed method has obvious advantages.Then,a river segmentation method for synthetic aperture radar(SAR)images is proposed,which is based on improved Chan-Vese(CV)model combining with reciprocal gray entropy multi-threshold selection optimized by artificial bee colony algorithm.Considering the uniformity of the gray level within river object cluster and background cluster,a coarse river image segmentation is made by using the multi-threshold selection algorithm based on reciprocal gray entropy and artificial bee colony optimization;Contrapose the low convergence speed and the sensitivity to initial conditions of basic CV model,the Dirac function is replaced with the image edge intensity and the coarse segmentation results serve as the initial condition of improved CV model which is utilized to make a fine segmentation for the river image.A large number of experimental results show that,the proposed segmentation method needs not set initial conditions and has high running speed as well as segmentation accuracy.Next,a method of target edge extraction for remote sensing images is put forward based on improved mathematical morphology and modulus maxima of NSST.Firstly,the image is decomposed into high-frequency components with more edges and details and low-frequency component with less edges and minutiae through NSST.Then considering the property of coefficients of edge points under different decomposing conditions,the modulus maximum detection is performed for each sub-band of high-frequency components and the double-layer mask is adopted afterwards so as to get the high-frequency edge extraction result.Moreover,the low-frequency component is processed through the improved mathematical morphology method to get the low-frequency edge extraction result.Finally,the above two parts are fused and the final target edge image is obtained after removing the isolated points according to the regional connectivity.A large number of experimental results show that,compared with Canny method and several recent similar edge extraction methods,the detected edges by the proposed method are accurate,clear,complete and with abundant details.It has strong anti-noise performance.Subsequently,a remote sensing image matching method in wavelet domain is explored based on Krawtchouk invariant moment and cuckoo search algorithm.Firstly,the wavelet transform is conducted for the reference image and the image to be matched respectively.Then low-frequency components upon the highest scale of these two remote sensing images are matched.Finally,according to the above result,the matching between the higher resolution images could be implemented stepwise up to the full resolution images.During the matching process,the Krawtchouk invariant moment is taken as matching feature and the cuckoo search algorithm is introduced as searching strategy to find the optimal matching point so as to realize the remote sensing image matching.A large number of experimental results demonstrate that,compared with those of other four existing image matching methods,the proposed method has advantages in matching accuracy and speed.It also has a stronger anti-noise ability and rotating robustness.Finally,a remote sensing image fusion method on the basis of graph based visual saliency(GBVS)model and guided filtering is proposed.Firstly,IHS transform is performed on the multi-spectral image to obtain the I component.The average filtering is utilized for the obtained I component and the original panchromatic image to get their smoothing images and detail images.Then,the initial weighted maps of the panchromatic image and I component are achieved by saliency map comparision of these two images using the simplified GBVS model.In addition,the guided filtering is adopted to get smoothing weighted maps and detail weighted maps by optimizing initial weighted maps.The smoothing information and the detail information of these two images are respectively fused according to the above weighted maps.Finally,the new I component is obtained by weighted fusion of fused smoothing images and fused detailed images.The final fusion result is ultimately achieved by inverse IHS transform.A large number of experimental results show that,compared with the IHS method and the other four remote sensing image fusion methods put forward in recent years,the proposed method has the most abundant spatial information as well as the optimal spectral quality.It is superior to other methods in subjective visual effects and objective quantitative evaluation index such as spatial frequency,correlation coefficient,running speed and so on.