Study On Target Feature Extraction Technology Based On Infrared Spectrum

For the detection of long-distance indistinguishable point target,it usually occupies only a few pixels on the image plane;without the shape and texture information of the target,it is difficult to extract the features of the target only by the gray information and time sequence information of a few pixels,and it is not easy to accurately sense the target.Infrared multispectral measurement of long-range point target can increase the amount of information of the detected target in the spectral dimension.It is an important non imaging detection technology of point target and a necessary supplement to the imaging detection of long-range target.The infrared spectrum of target is mixed with the information of surface temperature,effective radiation area and emissivity of surface material,which brings the possibility of feature extraction and recognition of point target.It is easy to obtain the equivalent temperature and area,which can roughly estimate the temperature and area of the long-range point target,and can further mine the precise features of the target from the infrared spectrum,so it has an important application prospect in enhancing and expanding the perception ability of the long-range target.This paper focuses on the target feature extraction technology based on infrared spectrum,and the main contents are as follows.(1)A simulation model of infrared spectral radiation characteristics of target is established,in which the spectral emissivity of target surface materials and the change of target motion attitude are considered.And,it is mainly divided into five modules:vector coordinate transformation,earth angle coefficient,heat flux in orbit,target transient temperature field and infrared spectral radiation intensity of target in detection direction.The model can be used to calculate the infrared spectral radiation intensity of targets with different motion attitude in any ground-detection direction,and the infrared spectrum can show the emission characteristics of target surface materials,which lays a foundation for the analysis of infrared spectral radiation characteristics of targets.(2)Through the established model,the infrared spectral radiation characteristics of the target are analyzed in detail.Firstly,the typical three-axis stabilized target in low orbit is analyzed,and the commonness of infrared spectral radiation characteristics of the target is mastered.It is found that there are some undulating characteristics in the infrared spectral curve of the target,which are consistent with the characteristics of the main emission materials of the target.In addition,the change of the target radiation intensity is mainly caused by the target motion.Then,the infrared radiation characteristics of targets with different motion attitude are analyzed.It is found that the infrared spectral emission characteristics of targets with different materials are different.And,the time-series radiation intensity of targets with different motion attitude is quite different,which brings the possibility of target attitude identification.(3)The model of infrared spectral radiation measurement is established.The process of obtaining the numerical signal by the infrared radiation measurement system is described.The process of retrieving infrared spectral irradiance from numerical signal is described.The error of the infrared spectrum irradiance is analyzed.The overall error is as high as 20%,which has completely obscured the material emission characteristics in the target infrared spectrum.(4)the extraction of equivalent temperature and equivalent area which can be achieved under large measurement error is studied.Firstly,the extraction algorithm is studied.The extraction mechanism of equivalent temperature and equivalent area is described,and the improved Gauss Newton algorithm with stronger robustness,wider applicability and higher efficiency is proposed.Then the algorithm is used to extract the equivalent temperature and equivalent area of the simulation targets,and their application is studied.It is clear that the equivalent temperature can roughly estimate the temperature level of the target,and the equivalent area can represent the change of the target projection area.In addition,there are obvious differences in the equivalent temperature and equivalent area of different motion attitude targets.Using this difference,the classification and identification method of different moving attitude targets are studied.The result shows that the overall recognition accuracy can reach88.5% without noise,but it is greatly affected by the measurement noise.Finally,two methods are used to improve the accuracy of the equivalent temperature and area.One is the selection of infrared multi bands based on the improved Gauss Newton method,and the other is the extended Kalman filter algorithm considering the correlation information between data frames.The results show that both methods can effectively improve the accuracy of infrared rough features.(5)The infrared precise feature extraction of the point target is studied.The target precise feature extraction method which obtains the main material of the target as a priori condition is proposed.The simulation of the triaxial stabilized target shows that it can accurately obtain the material composition of the target,and the extracted temperature and area can also match the temperature and area of each part of the target.Error analysis shows that for the extraction of main material properties,the error needs to be within 6%,and for the extraction of precise features of each part of the target,the error needs to be within 5%.In addition,the experiment is carried out to verify the feasibility of target precise feature extraction.