Research On Target Detection Technology Of Hyperspectral Image Based On FPGA

In recent decades,the hyperspectral images obtained by remote sensing image systems have provided users with considerable information about ground features in different scenarios.The application potential of hyperspectral image processing technology in various fields such as atmospheric observation,disaster relief and military defense has also been developed.The solution of remote sensing images has been greatly improved.The amount of information in hyperspectral images has also grown explosively.The demand for real-time data in scientific research and practical applications has increased significantly,but the traditional way that users receive data on the spaceborne remote sensing data sensor and then transmit data to the user terminal for decompression and analysis cannot meet the requirements.The low power consumption,small size,radiation resistance,and high reconfigurability of FPGA(Field Programmable Gate Array)present new ideas for real-time on-board processing of remote sensing images.This article starts from the target recognition of hyperspectral remote sensing images.According to the ACE(Adaptive Coherence Estimator)target detection algorithm and RXD(Reed-Xiaoli Detector)anomaly detection algorithm based on background information statistics from remote sensing images,this article takes into account the calculation accuracy and design complexity of the algorithm,uses FPGA as a hardware system implementation platform and constructs a dual-mode hyperspectral image real-time target detection system that can both do target detection and anomaly detection.The main work of this paper is as follows.(1)This paper researches the main processing of hyperspectral image target detection,analyzes the detection algorithm principle and model,selects ACE target detection algorithm and RXD anomaly detection algorithm,and optimizes their key steps.(2)Based on different raw data and calculation speed,this paper analyzes four background statistics methods and two background autocorrelation matrix inversion methods that are closely related to computing performance.Then this paper matches more suitable optimization methods for different original images and detection requirements,reduces the computational complexity and facilitates the implementation on FPGAs.(3)Starting from the application of dual-mode real-time detection,the entire system is modularized and reasonably divided according to functions and let each system modules can calculate parallel,which further can improve the system's computing efficiency.It can implement target detection and anomaly detection functions based on different input raw data information,and can speed up calculating while maintaining high target detection accuracy.Finally,a complete hyperspectral image real-time target detection system based on FPGA is established.