Real-time Target And Anormy Detection Algorithm For Hyperspectral Imagery

Because of the high spectral resolution,a hyperspectral imaging sensor is now capable of uncovering the subtle signal sources that cannot be detected by traditional remote sensing sensors.Hyperspectral detection can be divided into target detection and anomaly detection according to different inputs of prior knowledge.As a result,hyperspectral detection has received considerable interest in military reconnaissance,pollution monitoring,mineral exploration and other areas.It is particularly crucial when target such as moving targets may appear in a short period and vanish thereafter,in which case,timely detection is necessary and real-time processing of anomaly detection becomes inevitable.This thesis addresses the real-time issues and further designs three types of real-time detectors based on the classic detection theory.The major works of the thesis is as follows:(1)In order to improve the real-time capability of target detection,we need to study the method to update the background information efficiently.Three detectors called adaptive matched filtering(AMF),adaptive cosine agreement estimator(ACE)and ellipse profile distribution model(ECD)have been widely used by hyperspectral imaging sensors.A causal sample covariance matrix is derived for data samples.As for the real-time detection,the identity is used in recursive update equation,which could avoid the calculation of historical information and thus speed up the processing.Based on these techniques,we can derive the recursive versions of the classic algorithms and achieve the real-time detection.Additionally,in this thesis,we also designed a real-time detection system by using the RT-AMF method.(2)Anomaly RX detection generally requires real-time processing to find targets on a timely basis.A novel non-causal real-time RX detection method named as LRT-K-RXD is proposed for hyperspectral anomaly detection,it is based on the sliding array window.When the data is received pixel by pixel,the center pixel of the sliding array window,which contains the new received data,is detected at the same time.The inverse operation of the local covariance matrix is replaced by the simple matrix multiplication,matrix addition and subtraction using the identity,which can reduce the computational cost and storage burden significantly,And the proposed LRT-K-RXD can meet the real-time requirement without reducing the detection accuracy.(3)This thesis also develops a recursive local summation RX anomaly detection(R-LS-RXD)approach by the virtue of sliding windows.A causal sample covariance/correlation matrix is derived for local window background.As for the real-time sliding windows,the identity is used in recursive update equations,which could avoid the calculation of historical information and thus speed up the processing.Furthermore,a background suppression algorithm(R-BS-LS-RXD)is also proposed in this thesis,which removes the currently testing pixels from the recursive update processing.Experiments are implemented on simulated images and real hyperspectral images.The experimental results demonstrate that the proposed detectors outperforms the traditional detectors.Compared with the traditional detectors,the proposed detectors can reduce the calculation time and storage space significantly.