| High-resolution remote sensing satellites have certain width limitations.When conducting large-scale ground observation applications,a single remote sensing image cannot cover all the information of the target area.Usually,multiple high-resolution remote sensing images need to be stitched together.Due to the influence of shooting time and weather conditions,the images involved in the stitching process will have significant color differences,which will cause serious stitching marks in the final result.Additionally,the presence of thin clouds and fog can also cause problems such as misalignment of stitching boundaries.This thesis combines image processing algorithms to study the method of stitching high-resolution remote sensing images.The main research work and innovative points include:(1)Two algorithms for automatic generation of mosaic lines for remote sensing images are proposed.The first algorithm is based on energy values,where an energy function is set to assign corresponding energy values to each pixel.Then,using dynamic programming and constrained conditions,an optimal mosaic line is obtained.The second algorithm uses the human visual system to select the mosaic line.The overlapping area of the images is quantified using human visual system-related indicators,and a weight map is obtained.The edge reference information for selecting the mosaic line is then obtained using an edge detection algorithm.Finally,the optimal path algorithm is used to obtain the desired mosaic line.The use of the human visual system to determine the position of the seam line can effectively reduce the discontinuity in the mosaic results,and the presence of the mosaic line is hardly noticeable subjectively.(2)To address the issues of inconsistent image color and the presence of thin clouds and fog,this thesis performed weak light enhancement and cloud removal on the images prior to mosaicking.In the weak light enhancement stage,a global-local discriminative structure was used to orient the global and local information and avoid overexposure.Additionally,a selfregularized perceptual loss method was employed to constrain the feature distance between the weak light input image and the enhanced image,ensuring that image details were not damaged.In the thin cloud removal stage,a generative adversarial network with spatial attention was used to recognize and focus on cloud areas from local to global levels,achieving the effect of removing thin clouds and fog.Experimental results showed that the processed images effectively avoided the occurrence of mosaic lines crossing rivers,buildings,farmland,and other areas,and significantly reduced the traces of mosaicking.This approach can be widely applied in the field of high-resolution remote sensing image mosaicking.(3)A high-resolution remote sensing image mosaicking system for enterprises and users has been designed and developed.The system has basic functions such as image visualization,querying,downloading,dataset management,and user management.The system integrates mosaic algorithms and data processing algorithms,enabling algorithm operations to be performed on the system side,while experimental training and processing are carried out on the server side,and processing results are returned to the system side for feedback.This approach maximizes the value of algorithms in practical applications.Through the series of works described above,this thesis solves the problem of incomplete display of target area information in the practical use of remote sensing satellite images.By extracting mosaic lines to perform image mosaicking,the original information of the image is maximally retained,and the authenticity of the data is ensured.Weak light enhancement and thin cloud and fog removal are performed on the image data to ensure color consistency of the mosaicking result and improve its visual observability.Experiments have shown that the mosaicked images generated by this method are significantly better than those produced by remote sensing image processing software currently available on the market.Based on this method,the high-resolution remote sensing image mosaicking system designed in this thesis solves the problems of complex mosaicking processes and frequent manual intervention in current production processes,providing convenience for generating high-quality mosaicked images. |
