Research On Hyperspectral Band Fusion Method For Satellite-Ground Collaborative Data Processing

Due to the high data quality and limited susceptibility to complex geographical conditions and artificial events,satellite-based hyperspectral remote sensing has been widely used.However,the traditional satellite-based remote sensing data processing system has some problems,such as complex application mode,centralized data processing,mutual independence of each link,slow response speed,etc.,which cannot meet the demand for efficient processing of massive hyperspectral data.Data fusion technology is an important means of mining effective information from massive data,which is often used in hyperspectral remote sensing data processing system.According to different fusion levels,data fusion methods of remote sensing images can be divided into pixel level fusion,feature level fusion and decision level fusion.Pixel level fusion has low information loss and high processing accuracy,but poor realtime performance;Feature level fusion involves partial data compression,resulting in a certain degree of information loss;Decision level fusion has fast processing speed and low communication volume,but it has the drawbacks of large information loss and low processing accuracy.To reduce information loss and improve the speed and accuracy of data processing in remote sensing,this paper proposes a band fusion model based on satellite-ground collaborative data processing.Different hyperspectral band fusion methods are designed for different tasks,and a weighted iterative fusion optimization method is developed in conjunction with band evaluation criteria.The main research contents are as follows:(1)Space-ground collaborative data fusion processing model based on band level: To realize efficient interconnection between space and ground,a space-ground collaborative data processing model based on band fusion is proposed.Two fusion models are proposed according to different objects to be fused,namely,the satellite-station fusion mechanism that can sequentially fuse a single band,and the station-station fusion mechanism that can gradually fuse a subset of bands.Experiments show that the space-ground collaborative data processing model based on band fusion can realize the synchronous operation of data transmission and data processing,the application mode is simple and the response time is short,which can meet the demand for efficient processing of massive hyperspectral data.(2)Task-oriented hyperspectral real-time fusion processing methods: To solve the problem of high storage load and slow processing speed caused by massive hyperspectral data on remote sensing processing systems,first,objective functions were determined for different data analysis tasks.Based on the Woodbury matrix identity and block matrix inversion principle,fusion algorithms between single bands,fusion algorithms between band sets and single bands,and fusion algorithms between band sets were designed according to different fusion scenarios.Then,based on the RAD,CEM,and LSOSP algorithm,band fusion formulas were derived for target detection,anomaly detection,and spectral unmixing tasks.The band fusion method based on acquisition sequence,initial band driven,band priority and band selection are proposed.Experimental results show that the proposed method achieves synchronous processing of data transmission,band fusion and application analysis,reduces the requirement of satellite hardware configuration,and improves the processing efficiency of satellite-borne hyperspectral remote sensing data.(3)Weighted iterative fusion optimization method based on band evaluation criteria:Define band priority scores based on the orthogonal projection theorem,and then develop ranking-based band evaluation criteria and searching-based band evaluation criteria.Optimize band fusion methods by adjusting the order of bands.In the detection tasks,to make full use of the abundant spatial information inside the band,a weighted iterative real-time fusion algorithm combining global weight and local weight was constructed by combining the corresponding band evaluation criteria,which increased the difference between the ground object and the background pixel,and generated a band fusion optimization mechanism with forward iterative feedback.The experimental results show that the proposed algorithm enhances the representation ability of ground objects in the band,and has a good performance in both anomaly detection and target detection tasks.It improves the accuracy and efficiency of detection tasks,realizes the accurate fusion and dynamic analysis of hyperspectral data,and has a broad application prospect in spaceborne hyperspectral image detection tasks.