Subpixel Temperature Estimation And Application Based On Infrared /Hyperspectral Data

With the continuous development of science and technology,the spatial resolution of hyperspectral images has grown from 10 m/pixel to 1 m/pixel,and the spectral resolution has evolved from dozens of bands to hundreds of bands.All of this lays a good foundation for more accurate target detection.However,for some special cases,such as camouflage targets,underground hidden targets,etc.Because the current hyperspectral target detection technology does not have the ability to penetrate,it is difficult to detect it,and the temperature inform ation can only be detected by the thermal infrared band.However,due to its special imaging method,the thermal infrared band has a very low spatial resolution.At present,the highest resolution of the spaceborne sensor is nearly 50 m/pixel,and the signal-to-noise ratio is very low,which is seriously disturbed by noise.Therefore,under such resolution conditions,it is only capable of detecting large-scale targets such as forest fires and underground nuclear power plants,and does not have the ability to detect small strategic targets such as aircraft and tanks.Through our research,it is found that the near-infrared data contains both reflected energy and radiant energy.Therefore,this study improves the spatial resolution of thermal infrared data through the fusion of near-infrared and thermal infrared data to achieve sub-pixel temperature estimation.In addition,the auxiliary target detection technology detects small targets,and the specific contents are as follows:Firstly,this paper starts with the thermal infrared temperature inversion algorithm and introduces the basic structure of the temperature ratio radiance separation algorithm.Three temperature inversion algorithms were used to invert the temperature of the thermal infrared band data from 800 nm to 1200 nm,and the results were compared with accuracy and time.By comparing the inversion results with the real results,the temperature inversion algorithm with the highest accuracy and the shortest time is selected to lay the foundation for the subsequent experiments.Then,the process of correlating near-infrared data with thermal infrared data is introduced.How to estimate the temperature of image sub-pixels by near-infrared data is described.And three different correlation models are listed,and the scatter plot is drawn by using the estimated value and the true value,and then the fitting is performed.By comparing the running efficiency of the algorithm and the accuracy of the results,the optimal correlation model is selected.To lay the foundation for subsequent target detection experiments.Subsequently,this paper proposes an anomaly detection method based on im-age whitening technology and principal component analysis technology.By comparing and analyzing the change of detection rate of traditional anomaly detection and proposed algorithm under different false alarm rates,the superiority of this algorithm is verified.Finally,the super-pixel segmentation technology is introduced to segment the temperature data.By comparing the histogram curves of each segmented image,the possible target regions are screened out.On this basis,the false alarm in anomaly detection is removed with spectral information,and the target to be detected is obtained.Through quantitative analysis,the superiority of the target detection algorithm proposed in this paper in complex scenes is verified.