Research On Thermal Infrared Hyperspectral Anomaly Detection Method Based On Low-rank Prior

Thermal infrared hyperspectral anomaly detection refers to the technology that separates the target from the background without the prior spectral information of the target.Compared with visible and shortwave infrared hyperspectral target detection,thermal infrared hyperspectral image can provide object's thermal information and object's emissivity information,and can realize day and night observation,which is superior to visible light in the detection of minerals and gases.However,only satellite thermal infrared data are available at this stage.Due to the high price of thermal infrared hyperspectral imaging spectrometers,there is currently no publicly available thermal infrared hyperspectral data based on spectrometers.In addition,because thermal infrared research involves complex atmospheric correction and the inversion mechanism of surface parameters,the research on thermal infrared hyperspectral anomalous target detection based on surface parameters is still in the development stage.The emissivity information of ground features characterizes the physical properties of ground features,and its hyperspectral characteristics can be used to detect and identify targets.The emissivity spectrum values of the ground features are mostly concentrated between 0.8-1.Compared with the visible shortwave infrared band,the emissivity spectrum contrast between the ground features is lower.In addition,thermal infrared images are susceptible to interference from atmospheric and instrument noise during imaging,resulting in complex image noise and low signal-to-noise ratio,which further reduces the emissivity spectral contrast of ground objects,and noise will also obscure part of the spectral characteristics.Therefore,hyperspectral anomaly detection algorithms based on spectral differences are difficult to separate the anomalies and background in thermal infrared images.Based on the above,there are two main difficulties in the research of thermal infrared hyperspectral anomaly target detection:(1)The construction of thermal infrared hyperspectral image anomaly detection data set.(2)Break through the low emissivity spectrum contrast and large image noise,and establish an abnormal target detection model.In this paper,the Hypercam thermal infrared hyperspectral imaging spectrometer is used to collect thermal infrared hyperspectral data,and research is carried out on the problems of low emissivity spectral contrast and large image noise of thermal infrared hyperspectral images.The main research contents and innovations are as follows:(1)The characteristics of thermal infrared images are analyzed from the perspective of noise and emissivity spectrum.Three sets of airborne thermal infrared images and two sets of ground thermal infrared images are constructed;(2)Analyze and summarize the classic statistical-based hyperspectral anomaly detection algorithms.Focus on the analysis of anomaly detection algorithms based on low-rank sparse theory;(3)Based on the characteristics of thermal infrared hyperspectral images,a joint emissivity and low-rank segmentation anomaly detection method(Emissivity and a segmented low-rank prior anomaly detection,Ea SLRP)is proposed,which solves the problem that it is difficult to separate anomalous targets in the background of thermal infrared image due to low spectral contrast and high noise.And the effectiveness and advantages of the method are proven through five sets of thermal infrared hyperspectral image data sets.This research proposes the emissivity and low-rank segmentation anomaly detection method to realize the separation of anomalies and background in thermal infrared hyperspectral images.It has great significance in the field of thermal infrared hyperspectral anomaly detection and provides theoretical and application support for thermal infrared hyperspectral anomaly detection.