Research On Remote Sensing Image Fusion In Emergency Logistics Road Extraction

Emergency logistics refers to the unexpected events in response to serious natural disasters,public health emergencies,public security event and the military conflict and the material,personnel,capital requirements were emergency safeguard a special logistics activities.Remote sensing technology has become a common means of traffic information collection because of its fast speed,wide coverage and short renewal period.However,because of the limitation of the bandwidth of the sensor,in the process of remote sensing imaging,it is difficult to get the comprehensive information through the single sensor,and the images collected by multi sensors not only cause a large amount of data redundancy,which affects the transmission efficiency,but also are not conducive to the human or computer vision perception.So,it is very important to use the image fusion technology to integrate the multiple remote sensing images with complement and synthesized information into an image.Also,image fusion technology can provide support for the decision making of emergency logistics quickly.Image fusion can be divided into two types of spatial domain and frequency domain method.The spatial domain methods are fused in the pixel grayscale or color space of the original image directly,this methods has simple calculation and low computation complexity,but poor performance in image detail expression,the fusion result is unsatisfactory.The multi-scale decomposition course for the resource images in frequency domain fusion methods is similar to the human visual system multi-scale decomposition for images,with the result that the transform domain methods replace the spatial methods gradually.The multi-scale analysis methods are widely used in the transform methods.At first,we introduce several frequently used multi-scale analysis methods,including wavelet transform,Curvelet transform and Contourlet transform and their application in image fusion,then we analysis their characteristics and existing problems through several experiments.In this paper,we make studies about the problems of multi-scale analysis methods in the remote sensing image fusion,and put forward the improved method.The main innovations include the following aspects:1.In allusion to the problem that the multi-scale analysis methods lack of translation invariance,we propose a novel image fusion method based on sparse representation and the non-subsample Shearlet transform.In comparison to the Curvelet transform and the Contourlet transform,the Shearlet transform is more efficient and the non-subsample operation in the decomposition process can effectively overcome the pseudo-Gibbs artifacts that come from the source images incomplete registration.In the fusion rules,we take the sparse representation process on the non-sparse low frequency coefficients,thus we can better make use of the inherent characteristics of the low frequency sub-band to select the low frequency coefficients.In addition,we also utilize the low frequency sub-band as the training sample directly to learn the dictionary.As to the sparse high frequency coefficients,the improved sum-modified-Laplacian is adopted to fuse the coefficients.By comparing with other multi-scale analysis methods,the effectiveness of the proposed method is proved.2.There are serious spectral distortion phenomena when using the multi-scale analysis methods in the remote sensing image fusion,and the computational complexity is very high,not conducive to the real-time image processing.To overcome these problems,we propose another remote sensing image fusion method based on the adaptive IHS transform and multi-scale guided filter,in which the spatial detail injection model is the basis.In the proposed method,the intensity component is obtained adaptively from the up-sampled MS image firstly.And then the intensity component is used as the guidance image to guided filter the PAN image to get the detail information.In order to obtain more spatial detail,the multi-scale guided filter strategy is utilized.Finally,we combine the spatial detail with the improved edge detecting weighting matrix,and the result was injected to the each band of the up-sampled MS image to obtain the high resolution remote sensing image.The proposed fusion method not only has easy implementation,low computational complexity,but also achieves better fusion result comparing with other state-of-the-art fusion methods,so the proposed method is more suitable for practical applications.